Handbook of Industrial Membrane Technology

HANDBOOK OF
INDUSTRIAL MEMBRANE
TECHNOLOGY
Edited by
Mark C. Porter
Consultant
Pleasanton, California
Reprint Edition
El
"P
NOYES PUBLICATIONS
Westwood, New Jersey, U.S.A.
Copyright 01990 by Noyes Publications
No part of this book may be reproduced or utilized in
any form or by any means, electronic or mechanical,
including photocopying, recording or by any informa-
tion storage and retrieval system, without permission
in writing from the Publisher.
Library of Congress Catalog Card Number: 88-17878
ISBN O-8155-1205-8
Printed in the United States
Published in the United States of America by

Noyes Publications
Fairview Avenue, Westwood, New Jersey 07675
10987654
Library of Congress Cataloging-in-Publication Data
Handbook of industrial membrane technology
Bibliography: p.
Includes index.
1. Membranes (Technology) --Handbooks, manuals, etc.
I. Porter, Mark C.
TP159.M4H38 1988 880.2’842 88-17876
ISBN O-81 55-l 205-8
Contributors
Richard W. Baker Gabriele lorio

Membrane Technology and University of Naples
Research, Inc. Naples, Italy
Menlo Park, CA
James E. Kurz
lngo Blume Permea, Inc.
Membrane Technology and St. Louis, MO
Research, inc.
Menlo Park, CA Robert J. Petersen
Filmlec Corporation
John E. Cadotte Minneapolis, MN
FilmTec Corporation
Minneapolis, MN Mark C. Porter
Consultant
Gerard0 Catapano Pleasanton, CA
University of Naples
Naples, Italy Robert Rautenbach
Rhein Westfalen Technische
Thomas A. Davis Hochschule Aachen
Graver Water Aachen, West Germany
Union, NJ
Heiner Strathmann
Enrico Drioli Fraunhofer-lnstitut fur Grenzflachen-
University of Naples und Bioverfahrenstechnik
Naples, Italy Stuttgart, West Germany
A. Keith Fritzsche Richard G. Sudak

Ramicon, Inc. Separation Processes, Inc.
Woburn, MA Solana Beach, CA
xi
NOTICE
To the best of the Publisher’s knowledge the

information contained in this book is accu-
rate; however, the Publisher assumes no re-
sponsibility nor liability for errors or any
consequences arising from the use of the
information contained herein. Mention of
trade names or commercial products does
not constitute endorsement or recommen-
dation for use by the Publisher.
Final determination of the suitability of

any information, procedure, or product
for use contemplated by any user, and the
manner of that use, is the sole responsibility
of the user. The book is intended for in-
formational purposes only. Expert advice
should be obtained at all times when im-
plementation is being considered.
xii
Contents
1. SYNTHETIC MEMBRANES AND THEIR PREPARATION .......... .I

Heiner Strathmann
Introduction. ..................................... .I
Definition of a Membrane. .......................... .2
Fluxes and Driving Forces in Membrane Separation Processes ... .2
Discussion of Technical Relevant Synthetic Membranes
Methods of Their Preparation ......................... .4
Neutral Microporous Membranes ...................... .4
Symmetric Microporous Sintered Membranes ............ .4
Stretched Membranes. ........................... .6
Capillary Pore Membranes. ........................ .7
Symmetric Microporous Phase Inversion Membranes ....... .9
Symmetric Microporous Phase Inversion Membranes from
Inorganic Materials ............................ 11
Asymmetric Microporous Membranes .................. .I2
Preparation Procedures of Asymmetric Membranes ........ 13
The Formation Mechanism of Microporous Symmetric or
Asymmetric Membranes. ....................... .I3
The Phase Separation Process and Its Relation to the
Formation Mechanism of Microporous Membranes ........ .I4
Phenomenological Description of Phase Separation ........ 15
Mathematical Description of Phase Separation .......... .I8
General Observations Concerning Structures and
Properties of Phase Inversion Membranes. ............ .20
Rationalization of the Various Membrane Preparation
Procedures. ............................... .29
Homogeneous .........................
Membranes. .37
Homogeneous Polymer Membranes. ................. .37
Homogeneous Metal and Glass Membranes. ............ .37
Liquid Membranes. ............................ .38
...
XIII
xiv Contents
Ion-Exchange Membranes. ....................... .39

Composite Membranes. ........................... .45
Preparation Procedures of Composite Membranes ........ .45
Industrial Scale Membrane Production. ................. .49
Membrane Modules and Their Fabrication ............. .50
Membrane Manufacturing Equipment .................. .53
Future Developments. .............................. .56
References. ..................................... .56
2. MICROFILTRATION .................................. .61

Mark C. Porter
Introduction. .................................... .61
Membrane Structure and Fabrication .................... .62
Tortuous-Pore Membranes ......................... .64
Phase-Inversion Process ......................... .64
Stretching Process ............................. .64
Thermal-Phase-Inversion Process ................... .65
Capillary-Pore Membranes ......................... .66
Pore Size Determination. ............................ .70
Challenge Tests. ................................ .70
Bubble-Point Test. .............................. .71
Mercury Intrusion Test. ........................... .77
Other Methods ................................. .77
Flow Permeability Test. ......................... .77
SmokeDOPTest...............................7 8
BET Adsorption Data. .......................... .78
Retention Characteristics ............................ .78
Bacteria Retention and Bubble Point. .................. .78
Retention of Deformable Particles .................... .82
Retention by Adsorption .......................... .82
Retention by “Charged” Membranes. .................. .83
Aerosol Retention. .............................. .86
Direct Interception ............................ .88
Inertial Impaction. ............................ .88
Diffusional Deposition .......................... .88
Membrane Plugging and Throughput. .................... .90
Membrane Dirt-Loading Capacity. .................... .90
Prefilters ..................................... .90
Effect of Filtration Rate on Throughput ................ .95
Backwashing. .................................. .99
Cross-Flow Filtration. ............................ .99
Membrane Configuration ............................ 106
Plate and Frame Units ........................... .I06
Pleated Cartridges ............................... 106
Tubular/Hollow-Fiber Modules ...................... 113
Applications. .................................... 114
Through-Flow Filtration Applications. ................. 114
Sterilization and Particle Removal (Pharmaceuticals) ...... 114
Contents xv
Sterilization and Particle Removal (Beverages) .......... 117

Particle Removal (Semiconductor Process Fluids) ........ 119
Particle Removal (Nuclear Power Industry) ............ 124
Cross-Flow Filtration Applications. ................... 124
Removal of Heavy Metals ........................ 124
Industrial Laundry Wastewater. .................... 125
Plasmapheresis ............................... 126
Cell Harvesting/Washing ......................... 128
Continuous Cell Culture. ....................... .131
Prefiltration for UF ............................ 131
Membrane Distillation .......................... 132
Summary and Forecast. ............................. 134
References. ..................................... 134
3. ULTRAFILTRATION. ................................ .I36

Mark C. Porter
Introduction. ................................... .I36
Membrane Structure ....................
and Fabrication 138
Procedure for Casting Flat-Sheet Membranes ............. 140
Procedure for Casting Tubes ........................ 144
Casting Variables for Tubes and Sheets ................. 144
Procedure for Spinning Hollow Fibers. ................. 149
Spinning Variables for Hollow ..................
Fibers 152
Preparation of Inorganic Membranes. .................. 153
Dynamic Membranes .......................... .I53
Ceramic Membranes. ........................... 154
Pore Size Determination. ............................ 155
Molecular Weight Cut-off and Pore Size. ................ 155
Retention of Spherical and Linear Molecules ............. 157
Maximum Pore Size and Effective Pore Size .............. 158
Retention Characteristics ............................ 159
Adsorption Losses, .............................. 159
“Charged” Membranes. ........................... 161
Effect of Pressure ............................... 161
Fractionation of Solutes. .......................... 164
Membrane Flux with Concentration Polarization. ............ 166
Gel-Polarization ................................ 167
Effect of Pressure ............................. 167
Effect of Concentration ......................... 171
Effect of pH. ............................... .I73
Evaluation of the Mass-Transfer Coefficient. ............. 174
Laminar Flow. ............................... 174
Turbulent Flow. ............................. .I76
Theoretical Prediction of Flux ....................... 179
Macromolecular Solutions. ....................... 179
Colloidal Suspensions. .......................... 180
Tubular Pinch Effect ........................... 186
Augmented Cross-Flow Effects ...................... 192
xvi Contents
Centrifugal Force ............................. 192

Electric Field ................................ 192
Effect of Temperature ............................ 195
Membrane Fouling; Flux Decay and Restoration ............. 198
Effect of Cross-Flow Velocity. ...................... 199
Effect of Pressure .............................. .200
Effect of Membrane Surface Treatments ................ 200
Membrane Configuration ........................... .202
Tubes.......................................20 2
Hollow Fibers. ................................ .205
Plate and Frame Units ........................... .208
Spiral Wound Modules ........................... .212
UF Plant Design ................................. .214
Mode of Operation ............................. .214
Optimum Recirculation Rate. ....................... 216
Applications. ................................... .217
Ultrapure Water ............................... .218
Semiconductor Industry. ....................... .218
Pharmaceutical Industry. ........................ 221
Electrocoat Paint. .............................. .222
Oil-Water Separations ............................ 224
Reclamation of Waste Lubricating Oil .................. 226
Decontamination of Crude Oil. ..................... .227
PVA Recovery ................................ .227
Dyestuff Recovery and Purification ................... 229
Latex Concentration /Recovery. ..................... 229
Removal of Heavy Metals .......................... 230
Pulp and Paper Waste Treatment ..................... 230
Dairy Applications. ............................. .232
Cheese Whey Protein Recovery .................... 232
Milk Concentration ........................... .236
Food and Beverage Applications (Non-Dairy). ............ 237
Soy Whey ................................. .237
Egg White ................................. .237
Gelatin ................................... .238
Fruit Juice. ................................ .238
Wine......................................24 0
Beer......................................24 0
Corn Starch ................................ .241
Pharmaceutical and Biotechnology Applications ........... 241
Cell Harvesting. ............................. .242
Enzyme Concentration and Purification. .............. 242
Blood Plasma Processing. ........................ 242
Diafiltration. ............................... .243
Virus Concentration. ........................... 247
Enzyme Reactors ............................ .247
Membrane Fermentors ......................... .252
Waste Treatment. ............................ .254
Contents Xvii
Summary. ..................................... .255

References. .................................... .256
4. REVERSE OSMOSIS. ................................. .260

Richard G. Sudak
Introduction ................................... .260
Basic Process Considerations .......................... 263
Reverse Osmosis Membranes .......................... 270
Membrane Packaging .............................. .274
Plate and Frame ............................... .274
Spiral Wound ................................. .274
Tubular. .................................... .277
Hollow Fine Fiber. ............................. .279
Dynamic Membranes ............................ .281
Plant Design. ................................... .281
Pretreatment Section ............................ .283
Reverse Osmosis Section. .......................... 286
Posttreatment Section ............................ 290
Industrial Reverse Osmosis at a Refinery .................. 290
Reverse Osmosis and Ion Exchange. .................... .296
Reverse Osmosis and Pollution Control .................. .298
Reverse Osmosis and Seawater Desalination ................ 299
General Applications of Reverse Osmosis. ................. 302
Costs of Reverse Osmosis ........................... .303
Future Projections. ............................... .305
References. .................................... ,305
5. THIN FILM COMPOSITE REVERSE OSMOSIS MEMBRANES. ..... .307

Robert J. Petersen and John E. Cadotte
Cellulose Acetate Membranes. ......................... 309
Microporous Polysulfone Supports for Composite Membranes .... 311
NS-100 Composite Membrane. ....................... .314
PA-300 and RC-100 Membranes ...................... ,316
Other lnterfacial Membranes Based on Polymeric Amines ....... 318
Interfacial Polymerization with Monomeric Amines:
NS-300 Membrane .............................. .320
NF-40 Composite Membrane. ........................ .323
NTR-7250 Composite Membrane. ..................... .324
FT-30 Composite Membrane. ........................ .327
NF-50 Composite Membrane. ........................ .332
Mechanism of Interfacial Membrane Formation. ............. 332
Sulfonated Polymer Composites: NS-200 Membrane. ......... 333
PEC-1000 Membrane. ............................. .335
Sulfonated Polysulfone Membranes. .................... .338
Plasma Polymerization in Composite Membrane Fabrication ..... 340
Solrox 0200 Membrane ............................ .340
Miscellaneous Composite Reverse Osmosis Membranes. ........ 342
Advantages of the Composite Membrane Approach .......... .343
References. .................................... .344
Xviii Contents
6. PROCESS DESIGN AND OPTlMlZATlON ................... .349

Robert Rautenbach
Introduction-Mass Transport at the Membrane Surface. ....... .349
The Local Mass Transport. ........................ .349
Influence of the Asymmetric Structure of Membranes ....... 351
Change of Conditions Along the Membrane ............. .353
Module Concepts and Design. ........................ .354
The Hollow Fiber Module. ........................ .354
Cascades ...................................... .360
Definitions. .................................. .361
Cascades Without Reflux ......................... .364
Reflux Cascades ............................... .364
Constant Reflux ............................. .364
Constant Cut Rate, Variable Reflux ................. 364
The “Equilibrium Curve” ......................... .367
Complete Mixing of Feed and/or Permeate. ............ 367
Membrane Column ............................. .368
Processes. ..................................... .374
Seawater Desalination by RO. ....................... 374
Total Desalination of Brackish Water. .................. 376
Automotive Industry-Combination of UF, Cross Flow
Filtration and Evaporation for Recycling of Detergents
and Process Water. ............................ .382
Galvanic Industry-Treatment of Effluents .............. .384
Gas Permeation. ............................... .387
Pervaporation ................................. .390
References. .................................... .399
7. ENZYME MEMBRANE REACTORS AND MEMBRANE

FERMENTORS.....................................40 1
Enrico Drioli, Gabriele lorio and Gerard0 Ca tapano
Introduction. ................................... .401
List of Symbols. ................................. .404
Enzyme Membrane Reactors (EMR) .................... .409
Dynamic Enzyme Gel Layer Reactors. .................. .426
Membrane Segregated Enzyme Reactors ................. .439
Reactors with Enzymes Segregated in the Lumen of
Hollow Fibers ............................... .440
Reactors with Enzymes within the Pores of Asymmetric
Membranes..................................44 5
Tube and Shell Membrane Reactors with the Biocatalyst on
the Shell Side. ............................... .451
Perspectives .................................. .455
Membrane Bound Enzymes in Continuous-Flow Systems. ..... .456
Self-Cleaning Enzymatic Ultrafiltration Membranes. ........ 465
Membrane Fermentors ............................. .466
Cell Recycle Fermentors. ......................... .468
Membrane Segregated Fermentors ................... .473
Contents xix
Complex Fermentation Systems. .................... .475

References. .................................... .476
8. ELECTRODIALYSIS. ................................. .482

Thomas A. Davis
Introduction. ................................... .482
Ion-Exchange Membranes. ........................ .483
Membrane Types. .............................. .484
Heterogeneous .............................. .485
Homogeneous. .............................. .486
Membrane Properties ............................ .486
Water Transport ............................... .487
ED Stacks ..................................... .487
Design Considerations ........................... .487
Staging of ED Stacks .............................. .490
Applications of ED ............................... .490
Bipolar Membranes ............................... .494
Electrodes and Stack Power. ......................... .496
Electrode Reactions and Materials ................... .496
Electrode Isolation ............................. .498
Power Supplies................................ .500
Concentration Polarization ........................... 502
Membrane Fouling. ............................... ,507
CostofED......................................50 9
References. .................................... .509
9. COUPLED TRANSPORT MEMBRANES. .................... .511

Richard Baker and ingo Blume
Introduction. ................................... .511
History and Background ............................. 512
Counter Transport and Co-Transport .................... 515
Theory........................................51 7
Characteristics of Coupled Transport Membranes. ............ 520
Concentration Effects ............................ 521
Feed and Product Metal Ion Concentration Effects ......... 523
pH and Metal .........................
Ion Effects .525
Complexing Agent Effects. ......................... 526
Interfacial Reaction Rate Effects ..................... 531
Concentration Polarization Effects .................... 532
Coupled Transport Complexing Agents .................. .537
Membrane System Applications and Design ................ 541
Applications. ................................. .541
Copper Recovery. ............................ .541
Cobalt and Nickel Recovery ...................... 542
Uranium Recovery. ........................... .543
Electroplating Rinse Waters. ...................... 543
Copper Etchant Baths .......................... 545
Phenol and Ammonia Recovery. .................. .547
xx Contents
Supported Liquid Membranes Process Design ............ 547

Emulsion Liquid Membranes Process Design. ............ 552
The Future. ................................... .555
References. ................................... .555
10. THE SEPARATION OF GASES BY MEMBRANES. ............ .559

A. Keith Fritzsche and James E. Kurz
Introduction. .................................. .559
A Short History .............................. .559
Theory of Gas Transport in Membranes. ................. 561
Porous Membranes. ............................ .561
Non-Porous Membranes. ........................ .563
Rubbery Polymers. .......................... .564
Glassy Polymers .............................. .565
Engineering Aspects. ............... ............. 570
Preparation of Membranes for Gas Separation. ............ .574
Permea’s (Monsanto) Composite Membrane ............. 574
Dry Cellulose Acetate Membranes. ................... 578
Composite Polyimide Membranes. .................. .579
Polyolefin Membranes for Air Separation. .............. 581
Other Membranes ............................. .581
Membrane Testing and Evaluation .................. 582
Future. .................................... .582
Applications of Gas Separating Membranes. ............... 582
Hydrogen Recovery. ........................... .583
Carbon Dioxide Separation ....................... .588
Oxygen/Nitrogen Separation ...................... .589
Dehydration. ................................ .590
Other Separations .............................. 590
References. ................................... .591
INDEX..........,...................................594
Preface
This handbook emphasizes the use of synthetic membranes for separa-

tions involving industrial or municipal process streams. Little will be said con-
cerning the use of membranes in medical applications as in artificial kidneys
or for controlled drug release.
Most of the membrane processes are pressure driven. The notable excep-
tion to this is electrodialysis (ED) by which ions are separated under the in-
fluence of an electric field. In addition, the chapter on coupled transport covers
processes which are driven under the influence of a concentration gradient.
Pressure driven processes include microfiltration (MF), ultrafiltration (UF),
reverse osmosis (RO), pervaporation (PV), and gas separations (GS). With
the exception of pervaporation, which is just beginning to emerge as an in-
dustrial process, each of the above will be covered in a separate chapter.
The pressure driven liquid filtration processes (MF, UF, and RO) may be
distinguished by the size of the particle or molecule the membrane is capable of
retaining or passing. This roughly relates to the pore-size of the membrane.
Figure P.l shows the pore sizes of MF, UF and RO membranes. Obviously, all
particles or molecules larger than the rated pore size will be retained.
The smallest particle which can be seen with the naked eye, under the
best of lighting conditions, is about forty (40) microns in diameter. A typical
human hair has a diameter of eighty (80) microns. This means that membranes
are filtering particles out of solution which are invisible to the naked eye. Even
the most open MF membrane is capable of retaining yeast (3 to 12 microns)
and tighter MF membranes can retain the smallest bacteria (Pseudomonas
diminuta, 0.2 microns) (see Figure P.2).
The dividing line between MF and UF membranes at 0.1 microns (1000 8) is
somewhat arbitrary. The most open UF membranes (almost always anisotropic)
have a molecular weight cut-off of approximately one million (106) daltons
which corresponds to about 0.08 microns. There is some overlap in pore size
because there are MF membranes (usually isotropic) available with pore sizes
down to 0.02 microns. However, these MF membranes are used only for analyt-
ical applications and have no commercial importance for large scale processing.
V
vi Preface
o,oor)l UM 0.001 !JM 0,1 FIM 10 UM
Figure P.l: Pore sizes of RO, UF, and MF membranes.
vloleculor
Size Example Membrone process
weight
t 100pm Pollen
Starch
) 10Fm
Blood cells - Microf iltration
Typical bacteria -
Smallest bacteria -
) 1000
a
DNA, viruses
100,000
bl00a Albumin -
10,000 Ultrof iltrotion
1000 Vitamin 812 -

,lOR Glucose
Water _ Reverse osmosis

18 Na+CI- -
Figure P.2: Typical species retained by MF, UF and RO membranes.

Preface vii
Again, the dividing line between UF and RO membranes at 0.001 microns

(IO 8) is semantic. It is academic to even think of pores in RO membranes
since we are approaching the intermolecular spacing between the polymer
chains in the membrane. Usually, we think of the tighest UF membranes as
capable of retaining simple sugars and passing salts whereas RO membranes
would retain both.
Another distinguishing feature between UF and RO is that RO is usually
a high pressure process-operating between 200 and 1,200 psi (13 to 80 bar).
This is because the osmotic pressure of salt solutions can be quite high. Since
MF and UF membranes freely pass salt, there is no significant osmotic pressure
to overcome and they usually operate below 100 psi (7 bar).
The phenomenon of osmosis is schematically shown in Figure P.3. When
a salt solution is separated from pure water by a semipermeable membrane,
which allows no passage of salt, water diffuses through the membrane from
the higher chemical potential (pure water side) to the lower chemical potential
(salt solution side)-resulting in an increase in pressure on the salt solution side.
The increase above atmospheric pressure is called the “osmotic pressure.”
Obviously, to extract pure water from sea water, we must exceed the osmotic
pressure as in Figure P.4. Hence, the name “reverse-osmosis.”
Figure P.5 shows the osmotic pressure for several salts as a function of
their concentration. Sea water containing 3.5% NaCl will have an osmotic
pressure close to 400 psi (25 bar). In practice, pressures of 1,000 psi are used
in the reverse osmosis of sea water.
-T
Osmotic Head
(pressure)
e
Pure
Water
Semipermeable Membrane
Osmotic Equilibrium
Figure P.3: Osmosis.

viii Preface
Pressure.-
Semipermeable Membrane
Reverse Osmosis
Figure P.4: Reverse osmosis.
10,ooc
N
.c 1,OOC
1;
I
lO( I
IC)
0.1 1.0 10.0 100.0
Weight % solute
Figure P.5: Variation of osmotic pressure with salt concentration.

Preface ix
It has been estimated that the current worldwide membrane/equipment

market is 1.6 billion dollars annually (including hemodialysis). New applica-
tions in the emerging biotechnology field along with expanded use in waste
treatment promise rapid growth in the closing years of this century. It is hoped
that this volume will make a significant contribution to understanding the
potential and limitations of membrane technology.
Pleasanton, California M.C. Porter

October 1989 Editor
1
Synthetic Membranes and

Their Preparation
Heiner Strathmann
INTRODUCTION
In recent years, membranes and membrane separation techniques have grown

from a simple laboratory tool to an industrial process with considerable techni-
cal and commercial impact.’ Today, membranes are used on a large scale to pro-
duce potable water from the sea by reverse osmosis, to clean industrial effluents
and recover valuable constituents by electrodialysis, to fractionate macromolecu-
lar solutions in the food and drug industry by ultrafiltration,’ to remove urea
and other toxins from the blood stream by dialysis in an artificial kidney, and to
release drugs such as scopolamin, nitroglycerin, etc. at a predetermined rate in
medical treatment.j Although membrane processes may be very different in their
mode of operation, in the structures used as separating barriers, and in the driv-
ing forces used for the transport of the different chemical components, they have
several features in common which make them attractive as a separation tool. In
many cases, membrane processes are faster, more efficient and more economical
than conventional separation techniques. With membranes, the separation is usu-
ally performed at ambient temperature, thus allowing temperature-sensitive solu-
tions to be treated without the constituents being damaged or chemically al-
tered. This is important in the food and drug industry and in biotechnology
where temperature-sensitive products have to be processedB4Membranes can also
be “tailor-made” so that their properties can be adjusted to a specific separation
task.
Membrane science and technology is interdisciplinary, involving polymer
chemists to develop new membrane.structures; physical chemists and mathema-
ticians to describe the transport properties of different membranes using mathe-
matical models to predict their separation characteristics; and chemical engineers
to design separation processes for large scale industrial utilization. The most im-
portant element in a membrane process, however, is the membrane itself. To
1
2 Handbook of Industrial Membrane Technology
gain an understanding of the significance of the various structures used in

different separation processes a brief discussion of the basic properties and func-
tions of membranes, and the driving forces and fluxes involved is essential.
Definition of a Membrane
A precise and complete definition of a membrane which covers all its as-
pects is rather difficult, even when the discussion is limited to synthetic struc-
tures as in this outline. In the most general sense, a synthetic membrane is a bar-
rier which separates two phases and restricts the transport of various chemical
species in a rather specific manner.’ A membrane can be homogeneous or hetero-
geneous, symmetric or asymmetric in structure; it may be solid or liquid; it may
be neutral, may carry positive or negative charges, or may be bipolar. Its thick-
ness may vary between less than 100 nm to more than a centimeter. The elec-
trical resistance may vary from several megohms to a fraction of an ohm, Mass
transport through a membrane may be caused by convection or by diffusion of
individual molecules, induced by an electric field, or a concentration, pressure or
temperature gradient.
The term “membrane”, therefore, includes a great variety of materials and
structures, and a membrane can often be better described in terms of what it
does rather than what it is. Some materials, though not meant to be membranes,
show typical membrane properties, and in fact are membranes, e.g., protective
coatings, or packaging materials. All materials functioning as membranes have
one characteristic property in common: they restrict the passage of various
chemical species in a very specific manner.
Fluxes and Driving Forces in Membrane Separation Processes

Separations in membrane processes are the result of differences in the trans-
port rates of chemical species through the membrane interphase. The transport
rate is determined by the driving force or forces acting on the individual compo-
nents and their mobility and concentration within the interphase. The mobility
is primarily determined by the solute’s molecular size and the physical structure
of the interphase material, while the concentration of the solute in the inter-
phase is determined by chemical compatibility of the solute and the interphase
material, the solute’s size, and the membrane structure. The mobility and con-
centration of the solute within the interphase determine how large a flux is pro-
duced by a given driving force.
In membrane separation processes, there are three basic forms of mass trans-
port. The simplest form is the so-called “passive transport”. Here, the mem-
brane acts as a physical barrier through which all components are transported
under the driving force of a gradient in their electrochemical potential. Gradi-
ents in the electrochemical potential of a component in the membrane inter-
phase may be caused by differences in the hydrostatic pressure, the concentra-
tion, the temperature, or the electrical potential between the two phases sepa-
rated by the membrane. The second form of mass transport through the mem-
brane interphase is the so-called “facilitated” transport. Here, the driving force
for the transport of the various components is again the gradient in their electro-
chemical potential across the membrane. The different components, however,
are coupled to a specific carrier in the membrane phase. Facilitated transport,
Synthetic Membranes and Their Preparation 3
therefore, is a special form of the passive transport. Completely different, how-

ever, is the third form of mass transport through membranes. It is generally re-
ferred to as “active” transport. Here, various components may be transported
against the gradient of their electrochemical potential. The driving force for the
transport is provided by a chemical reaction within the membrane phase. Active
transport is coupled with a carrier in the membrane interphase and is found mainly
in the membranes of living cells.6 It has, to date, no significance in synthetic
membranes.
The transport of mass in a membrane is a nonequilibrium process and is
conventionally described by phenomenological equations such as Fick’s law
which relates the fluxes of matter to the corresponding driving forces, i.e., a
concentration gradient. The constant of proportionality is the diffusion coeffi-
cient. Driving forces in some membrane processes may be interdependent, giving
rise to new effects. Thus, a concentration gradient across a membrane may re-
sult in not only a flux of matter, but, under certain conditions, also in the buildup
of a hydrostatic pressure difference; this phenomenon is called osmosis. Similarly,
a gradient in hydrostatic pressure may lead to a concentration gradient as well
as a volume flow through the membrane; this phenomenon is called reverse os-
mosis. Frequently, fluxes of individual components are coupled, i.e., the flow of
one component causes a flow of another.’ One example of the coupling of fluxes
is the transport of bound water with an ion which is driven across a membrane
by an electrical potential gradient.
For membrane separation processes, only driving forces which induce a sig-
nificant flux of matter are of practical importance. These driving forces are hy-
drostatic pressure, concentration, and electrical potential differences. These
driving forces can also lead to the separation of chemical species.
(a) A hydrostatic pressure difference between two phases separated by

a membrane leads to a separation of chemical species when the hy-
drodynamic permeability of the membrane is different for differ-
ent components.
(b) A concentration difference between two phases separated by a
membrane leads to a separation of various chemical species when
the diffusivity and the concentration of the various chemical spe-
cies in the membrane are different for different components.
(c) A difference in the electrical potential between two phases sepa-
rated by a membrane leads to a separation of various chemical
species when the differently charged particles show different mobil-
ities and concentrations in the membrane.
The permeabilities of different components in a membrane depend on the

mechanism by which the components are transported. For example, in homoge-
neous polymer membranes, the various chemical species are transported under a
concentration or pressure gradient by diffusion. The permeability of these mem-
branes is determined by the diffusivities and concentrations of the various com-
ponents in the membrane matrix and the transport rates are, in general, rela-
tively slow. In porous membrane structures, however, mass is transported under
the driving force of a hydrostatic pressure difference via viscous flow and, in gen-
eral, permeabilities are significantly higher than in diffusion-controlled membrane

transport. In electrically charged membranes, usually referred to as ion-exchange
membranes, ions carrying the same charge as the membrane material are more or
less excluded from the membrane phase and, therefore, unable to penetrate the
membrane.s The type of membrane and driving force required for a certain mass
separation will depend on the specific properties of the chemical species in the
mixture.
For a given driving force, the flux through a unit membrane area is always
inversely proportional to the thickness of the selective barrier. For economic rea-
sons, membranes should in general be as thin as possible.
DISCUSSION OF TECHNICAL RELEVANT SYNTHETIC MEMBRANES AND

METHODS OF THEIR PREPARATION
Although synthetic membranes show a large variety in their physical struc-

ture and chemical nature, they can conveniently be classified in five basicgroups:
(1) microporous media, (2) homogeneous solid films, (3) asymmetric structures,
(4) electrically charged barriers and (5) liquid films with selective carriers. This
classification, however, is rather arbitrary and there are many structures which
would fit more than one of the abovementioned classes, e.g., a membrane may
be microporous, asymmetric in structure, and carry electrical charges. Any other
classification of synthetic membranes, e.g., according to their application or
methods of preparation, would serve the same purpose of phenomenologically
categorizing the various types of synthetic membranes.
Neutral Microporous Membranes

The neutral, microporous films represent a very simple form of a membrane
which closely resembles the conventional fiber filter as far as the mode of sepa-
ration and the mass transport are concerned. These membranes consist of a solid
matrix with defined holes or pores which have diameters ranging from less than
2 nm to more than 20 pm. Separation of the various chemical components is
achieved strictly by a sieving mechanism with the pore diameters and the par-
ticle sizes being the determining parameters. Microporous membranes can be
made from various materials, such as ceramics, graphite, metal or metal oxides,
and various polymers. Their structure may be symmetric, i.e., the pore diameters
do not vary over the membrane cross section, or they can be asymmetrically
structured, i.e., the pore diameters increase from one side of the membrane to
the other by a factor of 10 to 1,000. The properties and areas of application of
various microporous filters are summarized in Table 1 .I.
Symmetric Microporous Sintered Membranes. Sintered membranes are the
simplest in their function and in the way they are prepared. The structure of a
typical sintered membrane is shown in the scanning electron micrograph of Fig-
ure 1.1. This photograph shows a microporous membrane made out of poly-
tetrafluoroethylene by pressing a fine powder into, a film or plate of 100 to 500
pm thickness and then sintering the film at a temperature which is just below
the melting point of the polymer. This process yields a microporous structure of
relatively low porosity, in the range of IO to 40%. and a rather irregular pore
structure with a very wide pore size distribution.
Table 1.1: Microporous Membranes, Their Preparation and Application
Membrane type Membrane material Pore size Manufacturing process Application
Ceramic, metal 0.1 -20 pm pressIng and 51nterlng Mlcrollltration
or polymer powder of powder
Mlcroporous
membrane Homogeneous polymer 0.' -10 I'm stretching of extruded Microfiltration,
burn dressings,
sheets (PE, PTFE) polymer sheet.
artificial blood
vessels
Homogeneous polymer 0.02 -10 I'm track-etching Microfiltration
sheets (PC)
Polymer solution
0.01 -, "111 phase inversion Microfiltration.
ultrafiltration.
2nI11-'1'I11
(CN. CA) sterilization
Figure 1.1: SEM of a microporous sintered membrane prepared from a PTF E-

powder.
Sintered membranes are made on a fairly large scale from ceramic materials,
glass, graphite and metal powders such as stainless steel and tungsten.' The par-
ticle size of the powder is the main parameter determining the pore sizes of the
final membrane, which can be made in the form of discs, candles, or fine-bore
tubes. Sintered membranes are used for the filtration of colloidal solutions and
suspensions. This type of membrane is also marginally suitable for gas separa-
tion. It is widely used today for the separation of radioactive isotopes, especially
uranium.
Stretched Membranes. Another relatively simple procedure for preparing

microporous membranes is the stretching of a homogeneous polymer film of
partial crystallinity. This technique is mainly employed with films of polyethyl-
ene or polytetrafluoroethylene which have been extruded from a polymer pow-
der and then stretched perpendicular to the direction of extrusion.l0,11 This
leads to a partial fracture of the film and relatively uniform pores with diameters
of 1 to 20 .urn. A typical stretched membrane prepared from tetrafluoroethylene
is shown in the scanning electron micrograph of Figure 1.2.
Figure 1.2: SEM of a microporous membrane prepared by stretching an extruded

PTFE.film perpendicular to the direction of extrusion.
These membranes, which have a very high porosity, up to 90%, and a fairly
regular pore size are now widely used for microfiltration of acid and caustic so-
lutions, organic solvents, and hot gases. They have to a large extent replaced the
sintered materials used earlier in this application. Stretched membranes can be
produced as flat sheets as well as tubes and capillaries. The stretched membrane
made out of polytetrafluoroethylene is frequently used as a water repellent tex-
tile for clothing, such as parkas, tents, sleeping bags, etc. This membrane type
has, because of its very high porosity, a high permeability for gases and vapors,
but, because of the hydrophobic nature of the basic polymer, is up to a certain
hydrostatic pressure completely impermeable to aqueous solutions. Thus, the
membrane is repellent to rain water but permits the water vapor from the body
to permeate. More recently, this membrane has also been used for a novel proc-
ess, generally referred to as membrane distillation, i.e., to remove ethanol from
fermentation broths or wine and beer to produce low alcohol products” and for
desalination of seawater. These membranes are also used for desalination of sa-
line solutions and in medical applications such as burn dressings and artificial
blood vessels.
Capillary Pore Membranes. Microporous membranes with very uniform,
nearly perfectly round cylindrical pores are obtained by a process generally re-
ferred to as track-etching.13 The membranes are made in a two step process. Dur-
ing the first step, a homogeneous 10 to 15 pm thick polymer film is exposed to
collimated, charged particles in a nuclear reactor. As particles pass through the
film, they leave sensitized tracks where the chemical bonds in the polymer back-
bone we broken. In the second step, the irradiated film is placed in an etching
bath. In this bath, the damaged material along the tracks is profcrentially etched
forming uniform cylindrical pores. The entire process is schematically shown in
Figure 1.3. The pore density of a track-etched membrane is determined by the
residence time in the irradiator, while the pore diameter is controlled by the res-
idence time in the etching bath. The minimum port diameter of theso mcmbrancs
is approximately 0.01 pm. The maximum pore size that can bc achieved in track-
etched membranes is determined by the etching procedure. The polymer will not
only be dissolved along the sensitized track left by tho penetrating particle but
also on both surfaces of the film. Thus, with exposure time in the etching me-
dium the pore sizes increase and the thickness of the film is correspondingly re-
duced. The scanning electron micrograph in Figure 1.4 shows a typical track-
etched polycarbonate membrane. Capillary pore membranes are prepared today
mainly from polycarbonato and polyester films. The advantage of these poly-
mers is that they are commercially availabla in very uniform films of 10 to
15 pm thickness which is the maximum penetration depth of collimated par-
ticles obtained from a nuclear reactor which have an oncrgy about 0.8 to 1 MeV.
Particles with higher enorgy, up to IO McV, may be obtained in an accelerator.
They are used today to irradiate thickor polymer films, up to 50flm thickness,
or inorganic materials such as mica.14 Wowever, these mumbrancs arc not yet
available on a commercial basis.
Irradiation E tchm
1 c
t2
- etching
Figure 1.3: Schematic diagram illustrating the preparation of capillary pore

membranes by a track-etching process.
Figure 1.4: SEM of the surface of a capillary pore polycarbonate membrane.
Because of their narrow pore size distribution and low tendency to plug, cap-
illary pore membranes made from polycarbonate and polyester have found ap-
plication on a large scale in analytical chemistry and microbiological laborato-
ries, and in medical diagnostic procedures.15 On an industrial scale, capillary pore
membranes are used for the production of ultrapure water for the electronic in-
dustry. Here, they show certain advantages over other membrane products be-
cause of their short "rinse down" time and good long-term flux stability. Be-
cause of their surface filter characteristics, particles retained by the membranes
can be further monitored by optical or scanning electron microscopy. Figure
1.5 shows a scanning electron micrograph of asbestos fibers accumulated on a
capillary pore membrane in an air pollution control application. The membranes
are also used in standard clinical tests for red blood cell deformability studies.
Figure 1.5: SEM of asbestos filter accumulated on the surface of a capillary

pore membrane .
Human red blood cells have a diameter of approximately 6 to 8 Jlm. The human
body, however, contains capillaries approximately 3 Jlm in diameter. To pass
through these vessels the blood cells have to deform correspondingly. Healthy
cells will do this readily but malignant cells will not. By filtering blood through
a 3 Jlm capillary pore membrane certain blood deficiencies can be monitored.16
Symmetric Microporous Phase Inversion Membranes. The most important
commercially available, symmetric, microporous membranes are prepared by
the so-called phase inversion process.17 In this process, a polymer is dissolved in
an appropriate solvent and spread as a 20 to 200 Jlm thick film. A precipitant
such as water is added to this liquid film from the vapor phase, causing separa-
tion of the homogeneous polymer solution into a solid polymer and a liquid sol-
vent phase. The precipitated polymer forms a porous structure containing a net-
work of more or less uniform pores. A microporous cellulosic membrane made
by phase inversion is shown in the scanning electron micrograph of Figure 1.6.
This type of membrane can be made from almost any polymer which is soluble
in an appropriate solvent and can be precipitated in a nonsolvent.18 By varying
the polymer, the polymer concentration, the precipitation medium, and the pre-
cipitation temperature, microporous phase inversion membranes can be made
with a very large variety of pore sizes (from less than 0.1 to more than 20 Jlm)
with varying chemical, thermal, and mechanical properties. These membranes
were originally prepared from cellulosic polymers by precipitation at room
temperature in an atmosphere of approximately 100% relative humidity.19
Lately, symmetric microporous membranes are also prepared from Nylon 66,
Nomex, polysulfone, and polyvinylidene difluoride by precipitation of a cast
polymer solution in aqueous liquid.20
Figure 1.6: SEM of the surface of a microporous cellulose nitrate membrane

prepared by precipitation from a homogeneous polymer solution by water vapor
precipitation.
Polypropylene or polyethylene can also be used for the preparation of mi-

croporous membranes. However, since these polymers are not readily dissolved
at room temperature, the preparation technique must be slightly varied. Poly-
propylene, e.g., is dissolved in an appropriate amine at elevated temperatures. A
solution of 20 to 30% polymer is spread at elevated temperature into a film. The
precipitation of the polymer, however, is not induced by the addition of a non.
solvent but merely by cooling the solution to a point where a two phase system
forms. The resulting open foam structure is shown in Figure 1.7. The pore size
depends on polymer concentration, solvent system, solution temperature, and
cooling rate. This membrane preparation technique is usually referred to as ther.
mal gelation?l It can be applied to metal alloys and glasses as well as polymer
solutions.'
Figure 1.7: SEM of the surface of a microporous polypropylene membrane

precipitated by thermal gelation from a hot homogeneous polymer solution.
The symmetric, microporous polymer membranes made by phase inversion

are widely used for separations on a laboratory and industrial scale?2 Typical
applications range from the clarification of turbid solutions to the removal of
bacteria or enzymes, the detection of pathological components, and the detoxi-
fication of blood in an artificial kidney. The separation mechanism is that of a
typical depth filter which traps the particles somewhere within the structure. I n
addition to the simple "sieving" effect, microporous phase inversion membranes
often show a high tendency of adsorption because of their extremely large in-
ternal surface. They are, therefore, particularly well suited when a complete re-
moval of components, such as viruses or bacteria is desired. They are suited for
immobilization of enzymes or even microorganisms to be used in modern bio-
technology. They are also widely used for culturing of microorganisms in water
quality control tests.23 In this test, the water is filtered through a membrane
which retains all microorganisms. The membrane is then placed on a nutrient
pad in an incubator for 24 to 48 hours. During this time, the microorganisms
grow into easily visible colonies. Grid marked membranes assist in counting the
colony density, which is an indication of water quality. The process is shown in
the schematic diagram and the photograph of Figure 1.8.
Figure 1.8: SEM of cultured microorganisms in wotcr quality control tests and
schematic diagram of the process.
Symmetric Microporous Phase Inversion Membranes from Inorganic Ma-

terials. Microporous membranes can also be prepared by phase inversion from
glass or metal alloys. The preparation procedure is relatively simple: two dif-
ferent types of glass are homogeneously mixed; then, one glass type is removed
by acid or base leaching.% Thus, a microporous structure is obtained with
well defined pore sizes in the range of a few angstroms to several nanometers.
Porous glass membranes can be made in various configurations, such as flat
sheet, tubes, and hollow fibers, which are of particular importance because
of their high surface area. Microporous metal membranes can be prepared from
metal alloys such as Ni-Al-Cr by subsequent leaching of one component. These
membranes have found their main application in gas separation processes. While
the microporous metal membranes are mainly used for gas separation tasks, mi-
croporous glass membranes have been used mainly for separation of liquid mix-
tures including desalination of sea and brackish waters, ultrafiltration of waste-
water, and in artificial kidneys for detoxification of blood streams.25
Asymmetric Microporous Membranes

The most important membrane used today in separation processes is com-
posed of a rather sophisticated asymmetric structure. In this membrane, the two
basic properties required of any membrane, i.e., high mass transport rates for
certain components and good mechanical strength, are physically separated. An
asymmetric membrane consists of a very thin (0.1 to 1 pm) selective skin layer
on a highly porous (100 to 200 pm) thick substructure, as indicated in the sche-
matic drawing of Figure 1.9, which shows the cross section of (a) a symmetric
and (b) an asymmetric membrane. The very thin skin represents the actual mem-
brane. Its separation characteristics are determined by the nature of the polymer
and the pore size while the mass transport rate is determined by the membrane
thickness, since the mass transport rate is inversely proportional to the thickness
of the actual barrier layer. The porous sublayer serves only as a support for the
thin and fragile skin and has little effect on separation characteristics or the mass
transfer rate of the membrane. Asymmetric membranes are used primarily in
pressure driven membrane processes, such as reverse osmosis,, ultrafiltration, or
gas separation, since here the unique properties in terms of high mass transfer
rates and good mechanical stability can be best utilized.26
a) b)
Figure 1.9: Schematic drawing of the cross section of a (a) symmetric and (b)
asymmetric membrane.
In addition to high filtration rates, asymmetric membranes are most fouling

resistant. Conventional symmetric structures act as depth filters and retain par-
ticles within their internal structure. These trapped particles plug the membrane
and the flux declines during use. Asymmetric membranes are surface filters and
retain all rejected materials at the surface where they can be removed by shear
forces applied by the feed solution moving parallel to the membrane surface.
The difference in the filtration behavior between a symmetric and an asymmetric
membrane is shown schematically in Figure 1.10. Two techniques are used to
prepare asymmetric membranes: one utilizes the phase inversion process and the
other leads to a composite structure by depositing an extremely thin polymer
film on a microporous substructure.
a)
bP
-- --_ -
Figure 1.10: Schematic diagram of the filtration characteristics of a (a) sym-
metric and (b) asymmetric membrane.
Preparation Procedures of Asymmetric Membranes. The development of

the first asymmetric phase inversion membranes was a major breakthrough in the
development of ultrafiltration and reverse osmosis. These membranes were made
from cellulose acetate and yielded fluxes 10 to 100 times higher than symmetric
structures with comparable separation characteristics. Asymmetric phase inver-
sion membranes can be prepared from cellulose acetate and many other poly-
mers by the following general preparation procedure:*’
(I) A polymer is dissolved in an appropriate solvent to form a solution

containing 10 to 30 weight % polymer.
(2) The solution is cast into a film of typically 100 to 500pm thickness.
(3) The film is quenched in a nonsolvent typically water or an aqueous
solution.
During the quenching process the homogeneous polymer solution separates

into two phases: a polymer-rich solid phase, which forms the membrane struc-
ture, and a solvent-rich liquid phase, which forms the liquid-filled membrane
pores. Generally, the pores at the film surface, where precipitation occurs first
and most rapidly, are much smaller than those in the interior or the bottom side
of the film, which leads to the asymmetric membrane structure. There are dif-
ferent variations to this general preparation procedure described in the literature;
e.g., Loeb and Sourirajan used an evaporation step to increase the polymer con-
centration in the surface of the cast polymer solution and an annealing step dur-
ing which the precipitated polymer film is exposed for a certain time period to
hot water of 70” to 80°C.*s
The Formation Mechanism of Microporous Symmetric or Asymmetric
Membranes. The original recipes and subsequent modifications for preparing
asymmetric membranes are deeply rooted in empiricism. Detailed descriptions of
membrane preparation techniques are given in the literature.2J~30 Only after ex-
tensive use of the scanning electron microscope, which provided the necessary
structural information, was it possible to rationalize the various parameters for
membrane preparation processes. At first, the asymmetric structure was ob-
tained mainly in membranes made from cellulose acetate. But later it became ap-
parent that the process of making skin-type asymmetric membranes by precipi-

tating a polymer solution is just a special case of a general procedure, which
Kesting*’ refers to as phase inversion. During a phase inversion process, a one-
phase polymer solution is converted into a two-phase system consisting of a
solid (polymer-rich) phase which forms the membrane structure and a liquid
(polymer-poor) phase which forms the pores in the final membrane,
Phase inversion membranes can be prepared from any polymer mixture
which forms, under certain conditions of temperature and composition, a homo-
geneous solution and separates at a different temperature or composition into
two phases. A further condition, however, is that both phases are continuous. If
the liquid phase is discontinuous, a closed-cell foam structure will be obtained
and if the solid phase is discontinuous, a polymer powder will be obtained in-
stead of a rigid structure.
The actual phase separation can be induced by several means. One technique
described in the literature is based on the controlled evaporation of a volatile
solvent from a three-component mixture of solvent/precipitant/polymer, causing
precipitation as the system becomes enriched in precipitant. This technique was
used by Zsigmondy3’ and more recently by Kesting3* for the preparation of ul-
trafiltration and reverse osmosis membranes. Alternatively, precipitation of a
simple two-component polymer-solvent casting solution can be brought about
by imbibing precipitant from the vapor phase. This technique was the basis of
the original microporous membranes and is still used commercially today by sev-
eral companies. Yet, another technique is to bring about precipitation by cooling
a casting solution which forms a homogeneous solution only at elevated tem-
perature, e.g., polypropylene dissolved in N,N-bis-(2-hydroxyethyl)tallow amine,
This process is called thermal gelation. Thermal gelation is not only applicable to
polymers which, because of their poor solubility, would otherwise be inaccessi-
ble to the phase inversion membrane preparation techniques; it can also be used
for making microporous membranes from glass mixtures and metal alloys in
combination with a leaching procedure, as indicated earlier.
The Phase Separation Process and Its Relation to the Formation Mechanism of
Microporous Membranes
Detailed recipes given in the literature for the preparation of microporous
structures from polymers, metals or glassesseem superficially very different. But
in all cases, the basic membrane formation mechanism is governed by similar
thermodynamic and kinetic parameters, such as the chemical potentials and dif-
fusivities of the individual components and the Gibb’s free energy of mixing of
the entire system.33 Identification and description of the phase separation proc-
ess is the key to understanding the membrane formation mechanism-a necessity
for optimizing membrane properties and structures. The techniques used to in-
duce the phase separation in a homogeneous solution for the preparation of mi-
croporous membranes can be related to three basic procedures:
(1) Thermogelation of a homogeneous solution of two or more com-

ponents;
(2) Evaporation of a volatile solvent from a homogeneous solution of
two or more components;
(3) Addition of a nonsolvent or nonsolvent mixture to a homogeneous

solution.
All three procedures may result in symmetric microporous structures or in

asymmetric structures with a more or less dense skin at one or both surfaces
suitable for reverse osmosis, ultrafiltration or microfiltration. The only thermo-
dynamic presumption for all three preparation procedures is that the free energy
of mixing of the polymer system under certain conditions of temperature and
composition is negative; that is, the system must have a miscibility gap over a de-
fined concentration and temperature range.34
Phenomenological Description of Phase Separation. Phase separation due to
thermal gelation, evaporation of solvent and addition of nonsolvent can be illus-
trated with the aid of the phase diagram of a polymer solution.
Thermogelation of a Two-Component Polymer Mixture. The simplest pro-
cedure for obtaining a microporous system is by thermogelation of a two com-
ponent mixture, which at sufficiently high temperature forms a homogeneous
solution for all compositions, but at a lower temperature shows a miscibility gap
over a wide range of compositions. This behavior is illustrated schematically in
Figure 1.11, which shows a phase diagram of a two component mixture of a
polymer and a solvent as a function of temperature.
Homogeneous
, Solution
Miscibility Gap
/
Y
\ - Composition \
Solid Phase \
Liquid Phasi
Figure 1 .ll: Schematic diagram showing the formation of a microporous system

by thermal gelation of a two-component mixture exhibiting a miscibility gap at
certain conditions of temperature and composition.
The points P and S represent the pure components, polymer and solvent re-
spectively, and points on the line P-S describe mixtures of these two compo-
nents. If a homogeneous mixture of the composition X, at a temperatureT,,
as indicated by the point A in Figure 1 .ll, is cooled to the temperature Tz, as

indicated by point 6, it will separate into two different phases, the composition
of which are indicated by the points B’ and 6”. The point 6’ represents the poly-
mer-rich solid phase and the point 6” the solvent-rich, polymer-poor liquid
phase. The lines B’-B and B”-B represent the ratio of the amounts of the two
phases in the mixture; that is, the overall porosity of the obtained microporous
system.
Evaporation of a Volatile Solvent from a Three-Component Polymer
Solution. This process utilizes a three-component mixture: a polymer, a volatile
solvent and a third component, which by itself is a nonsolvent for the polymer.
This three-component mixture is completely miscible over a certain composi-
tion range but exhibits a miscibility gap over another composition range, as in-
dicated in Figure 1.12, which represents an isothermal phase diagram of the
three components. The corners of the triangle represent the pure components.
Boundary lines between any two corners represent mixtures of two compo-
nents, and any point within the triangle represents a mixture of all three com-
ponents. Within a certain compositionally defined range of thermodynamic
states, all three components are completely miscible, whereas in a different
range-the miscibility gap-the system separates into two distinct phases. If the
volatile solvent is completely evaporated from a homogeneous mixture of 10%
polymer, 60% solvent and 30% nonsolvent, as indicated by point A in Figure
1.12, the composition of the mixture will change from that represented by
point A to that represented by point B. At point B, the system consists of only
two components: polymer and nonsolvent. Since this point is situated within
the miscibility gap, the system is separated into two phases: a polymer-rich
phase, indicated by point B’ forming the rigid structure, and the phase B”
forming the liquid filled pores of the membrane.
Polymer
Solvent Non-solvent
Figure 1.12: Schematic diagram showing the formation of a microporous system

by evaporation of a solvent from a three-component mixture exhibiting a
miscibility gap at certain conditions of temperature and composition.
Addition of a Nonsolvent to a Homogeneous Polymer Solution. This tech-

nique can again be rationalized with the aid of a three-component isothermal
phase diagram shown schematically in Figure 1.13. This phase diagram of the
three-component mixture exhibits a miscibility gap over a wide range of com-
positions. If a nonsolvent is added to a homogeneous solution consisting of poly-
mer and solvent, the composition of which is indicated by the point A on the
solvent-polymer line, and, if the solvent is removed from the mixture at about
the same rate as the nonsolvent enters, the composition of the mixture will
change following the line A-B. At point C, the composition of the system will
reach the miscibility gap and two separate phases will begin to form: a poly-
mer-rich phase represented by the upper boundary of the miscibility gap and a
polymer-poor phase represented by the lower boundary of the miscibility gap.
At a certain composition of the three-component mixtures, the polymer con-
centration in the polymer-rich phase will be high enough to be considered as
solid. This composition is represented by point D in Figure 1.13. At this point,
the membrane structure is more or less determined. Further exchange of solvent
and nonsolvent will lead to the final composition of the membrane, the porosity
of which is determined by point B. Point B represents a mixture of the solid
polymer-rich phase and the liquid solvent-rich phase as represented by points
B’ and B” respectively.
Polymer
Solvent Non-solvent
/
Figure 1.13: Schematic diagram showing the formation of a microporous system

by addition of a nonsolvent to a homogeneous polymer solution in a three-
component mixture exhibiting a miscibility gap at certain conditions of tempera-
ture and composition.
The description of the formation of microporous systems by means of the

phase diagrams, as illustrated in Figure 1.11 to 1.13, is based on the assump-
tion of thermodynamic equilibrium. It predicts under what conditions of tem-
perature and composition a system will separate into two phases and the ratio
of the two phases in the heterogeneous mixture, i.e., the overall porosity. How-
ever, no information is provided about the pore sizes, which are determined by
the spatial distribution of the two phases. Equilibrium thermodynamics is not
able to offer any explanation about structural variations within the membrane
cross section, that is, whether the membrane has a symmetric or asymmetric
structure or a dense skin at the surface. These parameters are determined by ki-
netic effects, which depend on system properties, such as the diffusivities of the
various components in the mixture, the viscosity of the solution and the chemi-
cal potential gradients which act as driving forces for diffusion of the various
components in the mixture. Because these parameters change continuously
during the phase separation, which constitutes the actual membrane formation
process, no transient states of equilibrium will be achieved. Especially in poiy-
mer systems, frozen states will often be obtained that are far from equilibrium
and that can be stable for long time periods. The chemical potential and dif-
fusivities of the various components in the system, and their dependencies on
composition, temperature, viscosity, etc., are difficult to determine by inde-
pendent experiments and, therefore, are not readily available. This makes a
quantitative description of the membrane formation mechanism nearly impos-
sible. A qualitative description, however, which allows rationalization of the
membrane formation and correlation of the various preparation parameters with
membrane structures and properties, is possible.35”9
Mathematical Description of Phase Separation. The thermodynamic state of
a system of two or more components with limited miscibility can be described in
terms of the free energy of mixing.40 At constant pressure and temperature, three
different states can be distinguished:
(1) A stable state where all components are miscible in a single phase.
In this state, the free energy of mixing is positive
AC>0 (P,T = const) (1)

(2) An unstable state where the homogeneous solution separates spon-
taneously into two phases which are then in equilibrium. In this
state, which is located within the miscibility gap, the free energy of
mixing is negative
AC<0 (2)
(3) An equilibrium state given by the phase boundary composition. In
this state, the free energy of mixing is zero
AG = 0 (3)
The free energy of mixing of a system describes the thermodynamic state

of the system and thus provides information about the system stability. If a sys-
tem is unstable and separates in two coexisting phases, transport of individual
components has to take place. The transport processes are determined by ther-
modynamic parameters, which are expressed by driving forces, and by kinetic
parameters, which are determined by diffusivities, i.e., the diffusion coefficient.
Fick’s law relates the diffusion coefficient to concentration gradients. However,
the actual driving forces for any mass transport are gradients in the chemical po-
tential of the individual components. The diffusion coefficients of individual

components can be related to their chemical potential driving force by:41
bPi
0. = Bi
I ( ) G PT
#
(4)
Here, Di is the diffusion coefficient of component i, I_ciis its chemical potential

and Xi its mol fraction. Bi is a mobility term, which is always positive.
The stable, unstable and equilibrium states of the system are not only char-
acterized by positive, negative or disappearing values of the free energy of mix-
ing but, since the diffusion coefficient is directly related to the chemical poten-
tial of the individual components, which again is a function of the Gibb’s free
energy, the state of a system is also characterized by:
(1) Stable state
6P 70 , . (Ix5
and oi; y > 0 (5)
( ax 1 T,P I ( 6 IllXi >
(2) Unstable state
< 0 and oi =. y
,
j . 6In
C InXi
< o (6)
(3) Equilibrium state
= 0 and pi =
Bi RT
7
I
, l 63
bltlXi
=o (7)
Here fis is an activity coefficient referring to the pure phase, R is the gas con-
stant and T is the absolute temperature, Bi is a mobility term and Xi the mol-
fraction of the component i.
Equations (4) to (7) indicate that in the case of spontaneous demixing the
diffusion coefficient will become negative, i.e., the individual components will
diffuse against their concentration gradients. Since:
Pi = Pp l PT lnXi l RTln ty (8)
6 RTInXi 6 RT In r:
(9)
6Xf l
dxi
= xy
RT 1 l 6 Inf:
6 InXi (IO)
= Pi ; (6G) PT = (Zi),, (II)

,
20 Handbook of industrial Membrane Technology
The chemical potentials of the individual components and their dependence

on concentration of other species in a mixture and the mobility term I3 and its
dependence on viscosity determine the phase separation process.
Many aspects of the formation of symmetric or asymmetric membranes can
be rationalized by applying the basic thermodynamic and kinetic relations of
phase separation. There are, however, other parameters-such as surface tension,
polymer relaxation, sol and gel structures-which are not directly related to the
thermodynamics of phase separation but which will have a strong effect on
membrane structures and properties. A mathematical treatment of the forma-
tion of porous structures is difficult. But many aspects of membrane structures
and the effect of various preparation parameters can be qualitatively inter-
preted.
General Observations Concerning Structures and Properties of Phase Inversion
Membranes. Before going into any detailed discussion of the formation mecha-
nism of microporous membranes, several general observations concerning the
membrane structure, preparation procedures, and mass transport properties are
described.
Typical Membrane Structures. Using scanning electron microscope tech-
niques, the four different structures shown in Figures 1.14 (a) to (d) can be ob-
served for phase inversion membranes. Photograph (a) shows a cross section of a
microfiltration membrane prepared from a cellulose nitrate solution by precipi-
tation in a humidity controlled environment. The membrane shows a “sponge”-
like structure with no skin on the bottom or top surface. Photograph (b) shows a
cross section of a typical ultrafiltration membrane prepared from a polyamide
solution and precipitated in water by immersing the polymer solution into a wa-
ter bath. The membrane shows a “finger-“-type structure with a dense skin at the
surface and large pores penetrating the entire membrane cross section. The pores
increase in diameter from the top to the bottom side. Photograph (c) shows the
cross section of a typical reverse osmosis membrane prepared from a polyamide
solution and precipitated in water by immersing the polymer solution into a wa-
ter bath. The membrane shows a sponge-type structure with a dense skin at the
surface and a porous structure underneath with increasing pore diameter from
the top to the bottom side of the membrane. The only differences in the prepa-
ration procedures of the membranes shown in photographs (b) and (c) are the
polymer concentration and the precipitation temperature. Photograph (d) shows
the cross section of a reverse osmosis membrane prepared from a polyamide so-
lution and precipitated in a water-solvent mixture. The membrane shows a
sponge-type structure with a dense skin at the surface and a porous structure
underneath with a relatively uniform pore size distribution over the entire cross
section. Membranes with this type of structure can usually be dried without
changing their mass transport properties. Looking at the fine structure of a mem-
brane very often two types of structures may be obtained as indicated in Figure
1 .I 5 which shows scanning electron micrographs of the structures of two phase
inversion membranes. Figure 1 .15 (a) shows a sponge-type structure consisting
of spherical cells. This structure is often obtained from casting solutions with
relatively high polymer content. Figure 1.15 (b) shows a structure, which in the
literature is called “nodular”. It consists of small polymer beads randomly fused
together. The structure is often obtained from casting solutions with low poly-
mer concentrations. Very often in asymmetric membranes the structure may

change from nodular to “cellular” type over the membrane cross section, as
shown in Figure 1 .I 6.
b)
c) d)
Figure 1.14: Scanning electron micrograph of membrane cross sections with

typical structures: (a) symmetric microporous membrane without a “skin”;
(b) asymmetric membrane with a “finger”-type structure and a dense skin at
the surface; (c) asymmetric membrane with a “sponge’‘-type structure, a dense
skin, and pore sizes increasing from the surface to the bottom side; (d) sym-
metric membrane with a sponge structure, a dense skin and a uniform pore size
distribution in the substructure.
Figure 1.15: Scanning electron micrographs of the fine structure of two phase
inversion membranes: (a) sponge-type structure consisting of spherical cells;
(b) nodular structure consisting of small polymer beads fused together to form a
randomly porous structure.
Figure 1.16: Scanning electron micrograph of a membrane showing a nodular

structure at the surface which converts into a cellular structure towards the
bottom side of the membrane.
The Effect of the Polymer and Polymer Concentration on Membrane Struc-

tures and Properties. The scanning electron micrograph of Figures 1.17 (a) to
(c) show the cross sections of three membranes with nearly identical structures
and ultrafiltration properties (listed in Table 1.2) prepared from three different
polymers (cellulose acetate, polyamide and polysulfone) by precipitation in a
water bath. The scanning electron micrographs of Figures 1.18 (a) to (c) show
the cross sections of membranes made from one polymer-solvent system (poly-
amide in DMAC) with different polymer concentrations. These membranes show
completely different structures and filtration properties which are listed in Table
1.3. The scanning electron micrographs of Figures 1 .I 7 and 1 .I8 and the data
listed in Tables 1.2 and 1.3 indicate that the same type of membrane can be pre-
pared from various polymers and that from one polymer-solvent system, various
types of membranes can be made.
The polymer concentration is a very significant parameter for tailoring a
membrane in terms of its structure and separation properties. This is demon-
strated in Figure 1.19 where the scanning electron micrographs of membranes
are shown which are prepared from casting solutions of different polyamide
concentrations in NMP. With increasing polymer concentration, the structure
changes from a typical finger structure (5% polyamide in the casting solution)
to a typical sponge structure (22% polyamide in the casting solution).
The flux and retention properties, the membrane porosity and the rate of
precipitation change in a corresponding pattern, as shown in Table 1.4. With
increasing polymer concentration, the permeability decreases and the retention
increases. The overall porosity also decreases and the gelation time increases with
increasing polymer concentration.
C)
Figure 1.17: Scanning electron micrographs of membrane crosssections prepared

from three different polymer-solvent systems by precipitation in water: (a) 12%
cellulose acetate in DMAc; (b) 12% polyamide in DMSO; (cl 12% polysulfone in
DMF.
b)

from different polyamide solution concentrations and by different precipitation
procedures: (a) 10% polyamide in DMAc precipitated by water vapor; (b) 10%
polyamide in DMAc precipitated in a water bath; (c) 22% polyamide in DMAc
precipitated in a water bath.
Table 1.2: Filtration Properties and Porosities of Membranes Prepared

From Different Polymer-Solvent Systems by Precipitation in Water
Polymer Filtration rate* Rejection+ Membrane
km/r) (%I porosity
y-globulin bov.albumin (%)
in DMAc 3.5 x 1K3 99 98 80
Polyamide
in DMSO 2.1 x 10-1 97 72 82
Polysulfone
in DMF 1.9 x 10-3 96 80 83
l The filtration rater were determined with DI-water at 2 bar and room
temperature
+ The rejections were determined at 2 bar and room temperature with
solution of I % y-.@,lobulin and bov. albumin
Table 1.3: Filtration Properties and Porosities of Membranes Prepared

From Polyamide-DMAcWater Systems
Polymer- Precipitant Porosity Filtration Rejection+ structure
concentra- (%I rate*
tion km/S)
IO % polya- water vapor 89 2 x IO” 0 % for symmetric
mide in DMAc at 2 bar Dextran LOO “sponge”-
tYP=
10 96 polya- water 79 1.8 Y 1O-3 8a 96 for asymmetric
mide in DMAc at 2 bar bov.albumin “finger”-
type
22 % polya- water 71 8 x IO-’ 99 % for asymmetric
mid+ in’ DMAc at 100 bar MgSO,, “SpOllge”-
type
l The filtration propertin were determined with deionized water and solu-
tions of I 96solids at room temperature
+ The molecular weight of Dextran 100 was approximately 2 million Dalton


from various polyamide concentrations in NMP by precipitation in water at
room temperature.
Table 1.4: Rate of Precipitation and Filtration Properties of Membranes

Prepared From Polyamide-NMP Solutions of Various Concentrations by
Precipitation in Water at Room Temperature
Polymer Rejection’
concentra- Cyto-. bov. Filtration Celation
tion MgSO4 chrome C albumin rate Porosity time++
(%) km/r x 10’) (vol.%) (5)
5x 0 0 10 56 91 32
IO x 0 43 84 32 85 40
IS I 8 92 100 9 81 52
18 xx 7s 100 100 18 79 83
20 xx 90 100 100 4 77 142
22 xx 98 LOO 100 1.6 76 212
+ the rejection was determined with solutions of I % solids
++ for entire membrane
x applied pressure 5 bar
xx applied pressure LOObar
Rate of Precipitation and Membrane Structure. Certain membrane struc-

tures can be correlated with the rate of precipitation. Very high precipitation
rates (short gelation times) always lead to a “finger” structure, while slow pre-
cipitation rates (long gelation times) lead to asymmetric membranes with a
“sponge” structure, and very slow precipitation rates very often lead to symmet-
ric membranes with no defined skin at the surface, a structure with a very uni-
form pore size distribution over the entire cross section of the membrane. If a
sponge-type structure is obtained, the pore diameters are inversely proportional
to the rate of precipitation. Higher precipitation rates lead to finer pore struc-
tures, and lower precipitation rates lead to coarser structures. The rate of precip-
itation and its relation to the membrane structure can be easily observed with an
optical microscope or by other means.42
The Effect of the Polymer-Solvent-Precipitant System on Membrane Struc-
ture and Properties. The effect of the polymer on membrane structure and
properties is closely related to the solvent used in the casting solution and the
precipitant. The solvent and the precipitant used in membrane preparation de-
termine both the activity coefficient of the polymer in the solvent-precipitant
mixture and the concentration of the polymer at the point of precipitation and
solidification. The polymer-solvent-precipitant interaction can be approximately
correlated in terms of the disparity of the solubility parameter of polymer and
solvent.
(1) The smaller the solubility parameter disparity of solvent and polymer,
the better the compatibility of solvent and polymer, and the slower the precipi-
tation of the polymer. Thus, the tendency for a change from a sponge to a finger
structure membrane increases with decreasing compatibility of solvent and poly-

mer. (2) The higher this disparity, the less compatible are polymer and precipi-
tant, and the faster will be the precipitation. The tendency to change from a
sponge to a finger structure will thus increase with decreasing compatibility of
polymer and precipitant.
This is demonstrated in the scanning electron micrograph of Figure 1.20
which shows the cross section of membranes prepared from a solution of 15%
polyamide in DMAc precipitated in water-glycerin mixtures of different compo-
sitions. Since the compatibility of the polymer with glycerin is slightly better
than that with the water, i .e., the disparity in the solubil ity parameter of poly-
mer-water is larger than of polymer-glycerin, the tendency of the membrane
structure to go from a finger to a sponge structure will increase with increasing
glycerin content in the precipitation bath. Additives to the casting solution or
the precipitation bath have a similar effect.36
80%Gly..rln
Figure 1.20: Scanning electron micrographs of membrane cross sections prepared

from 15% polyamide in DMAc precipitated in water-glycerin mixture at room
temperature.
Rationalization of the Various Membrane Preparation Procedures. To dis-

cuss the formation of membrane structures it is useful to describe the concentra-
tion profiles of the casting solution components in the precipitating casting
solution. The schematic diagram of Figure 1.21 shows the concentration pro-
files of polymer. solvent and precipitant at some intermediate time during the
precipitation of a polymer film cast on a glass plate.
Poinl 01 Poinl01
SkIn, 10lidllicllion, pr.cipitati"n
.~ ':'--v-J: ..~
, -" ,
, ' , ,
Pr.cipll.nl ' pr.clpll.l.d ,lluid ' unpr.clpll.l.d .glall pill.
., ' ,
, polym.r ,polym.r calling lolullon ,
.,
, ' ,
.,
.,
, , .
., .
...
, .
-~ Polymer: :
c B D C A
.2 1OOr- I I
-
IV
I I
.. I I
-SO
i ~ ~ J I
U I -1 --/
C
O .' , .
/
(.) .' , .
.., .
.' .
.' ' .
.Solvent .
, .:
's b
I
c
I
~
"
100!'-
I I ---
"
--~
,- -- I
50
~.~
~ : ~..
u ...
... , .
.:
, .
, ' ,
Preclplt.nt .
, ..
'8 b C A
I I
100~~
50
.I
~ ~II I I
~-~
Membrane cross-section
Figure 1.21: Schematic diagram showing the concentration profiles of polymer ,

solvent and precipitant through a precipitating membrane: ---total concentration;
-concentration in the polymer-rich phase; -.-concentration in the polymer-
poor phase; A Casting solution composition; B Membrane composition; C Point
of precipitation; D Point of solidification.
During precipitation, the casting solution can be divided into three layers.
Traveling from the glass plate towards the precipitation bath these layers are:
(1) The casting solution layer: This layer is closest to the glass plate,
and has a composition similar to the original casting solution A.
Little solvent has diffused out of and little precipitant has diffused
into the layer.
(2) The fluid polymer layer: This layer lies between the point of pre-
cipitation and the point of solidification. In this layer, the casting
solution divides into a polymer-rich and a polymer-poor phase. At
the point of first phase separation, the polymer-rich phase still con-
tains a high solvent concentration and a low precipitant concentra-
tion and is still liquid. The polymer nearest the precipitation bath
has been precipitated longer, has lost solvent and has gained pre-
cipitant; its viscosity is therefore higher. Thus, the viscosity of the
precipitated polymer climbs from the point of first separation until
it becomes almost solid at the point of gelation. During this time,
bulk movement of the precipitated polymer takes place to form
the matrix of the final membrane.
(3) The solid polymer layer: In this layer, the solid, polymer-rich phase
undergoes continuous desolvation. Shrinkage, or syneresis, of the
solid polymer accompanying this composition change produces
stresses in the polymer. Because the polymer is solid, these stresses
cannot be as easily relieved by bulk movement of polymer as in the
fluid polymer layer. Instead, the polymer structure either slowly
undergoes creep to relieve the stress, or, if the stress builds up too
rapidly to be dissipated by creep, the polymer matrix breaks at
weak spots.
Formation of Symmetric and Asymmetric Membrane Structures. Two dif-

ferent techniques have been employed for the precipitation of membranes from
a polymer casting solution. In the first method, the precipitant is introduced
from the vapor phase. In this case, the precipitation is slow, and a more or less
homogeneous structure is obtained without a dense skin on the top or bottom
side of the polymer film. This structure can be understood when the concen-
tration profiles of the polymer, the precipitant, and the solvent during the pre-
cipitation process are considered. The significant feature of the vapor-phase pre-
cipitation process is the slow diffusion of precipitant from the vapor phase ad-
jacent to the film surface into the polymer solution. Because this process is so
slow, the concentration profiles in the cast film are flat. The concentration pro-
files of the precipitant in the polymer film are shown schematically in Figure
1.22 for various times. Because of the flat concentration profiles of the precipi-
tant over the cross section of the casting solution, the membrane precipitates
rather slowly and in the same way over the entire film cross section. Therefore, a
randomly distributed polymer structure is obtained during precipitation, as in-
dicated in the scanning electron micrograph of Figure 1.6.
In the second membrane precipitation procedure, the precipitant is added to
the casting solution by immersing the cast polymer film in a bath of the precipi-
tation fluid. In this case, the precipitation is rapid, and an asymmetric membrane
structure is obtained. This structure and formation can again be understood by
considering the concentration profiles of the polymer, the solvent and the pre-
cipitant during the precipitation process. These profiles over the cross section
of the cast polymer film are shown schematically in Figure 1.23. The most
important features in immersion precipitation are the steep concentration and
activity gradients of all components found in the polymer solution close to
the polymer-precipitation medium interface.
I membrane structure at t = 4 ’
I
I I
I I
I
I
point of solidification
point of precipitation
water vapor poly met- film

Figure 1.22: Schematic diagram showing the concentration profiles of the pre-
cipitant in the casting solution at various times during the formation of a sym-
metric structure membrane.
I- precipitant -
I I
I membrane structure at t = 4 ,
I
water bath polymer film

Figure 1.23: Schematic diagram showing concentration profiles of the precipitant
in the casting solution of various times during the formation of an asymmetric
skin-type membrane.
When the cast polymer film is immersed into the precipitation bath, solvent
leaves and precipitant enters the film. At the film surface, the concentration of
the precipitant soon reaches a value resulting in phase separation. In the interior,
however, the precipitant concentration is still far below the limiting concentra-
tion for phase separation. Phase separation therefore occurs initially at the
surface of the film, where, due to the very steep gradient of the chemical po-
tential of the polymer, there is a net movement of the polymer perpendicu-
lar to the surface. This leads to an increase of the polymer concentration in the
surface layer. It is the concentrated surface layer which forms the skin of the
membrane. This skin also serves to hinder further transport of precipitant into
and solvent out of the casting solution. The skin thus becomes the rate-limiting
barrier for precipitant transport into the casting solution, and the concentration
profiles in the casting solution interior become less steep. Thus, once the precipi-
tated skin is formed, the same situation in the sublayer is obtained as in a mem-
brane precipitated from the vapor phase, and a structure with uniform randomly
distributed pores is formed.
Skin Type Membranes With “Sponge’- and ‘Finger”-Like Structures. In
skin-type membranes, the two characteristic structures shown in Figure 1 .I4 are
obtained. One is a sponge-like structure and the other is a finger-like substruc-
ture underneath the skin.
The formation of the sponge-structured membranes can also be rationalized
by the precipitation process described above. With finger-structured mem-
branes, the formation process is more complex and cannot entirely be de-
scribed by the thermodynamic and kinetic arguments of phase separation
processes.43
Other phenomena, such as syneresis, shrinkage, and stress relaxation in the
precipitated polymer, also play an important role. The formation of finger-struc-
tured membranes is conveniently divided into two steps: the initiation and the
propagation of fingers. The formation of the skin is identical with that of the
sponge-structured membranes. However, as a result of syneresis, shrinkage stress
in the solid polymer skin cannot be relieved by creep relaxation of the polymer
and the homogeneous layer ruptures. The points at which the skin has been
fractured form the initiation points for the growth of the fingers. Once a finger
has been initiated, shrinkage of the polymer causes it to propagate by draining
the freshly precipitated polymer at the bottom of the finger to the side of the
finger. This is schematically shown in Figure 1.24 which shows the growth of a
finger at various times. Within a finger, the exchange of solvent and precipitant
is much faster than through the unfractured skin, and the precipitation front
moves much faster within a finger than in the casting solution bypassed between
fingers. This solution is protected from immediate exposure to the precipitant
by a layer of precipitated polymer. Precipitation therefore occurs much slower
and a sponge-like structure is formed between the fingers.
Uniform and Graded Pore Structure of Skin-Type Membranes. In sponge-
structured, skin-type membranes, it is assumed that the diffusion of the precipi-
tant through the skin is the rate-limiting step, leading to a more or less flat con-
centration profile of the precipitant in the casting solution just beneath the skin.
Depending on the resistance of the skin to the flux of precipitant, the concen-
tration profile may vary from a completely flat one (virtually no concentration
gradient over the cross section of the cast polymer film) to a concentration pro-
file showing an initially steep gradient at the beginning of precipitation which
decreases as precipitation proceeds through the polymer film. A uniform pore
structure is obtained when the concentration profiles are flat and the time that
the system needs to move from the point of precipitation to the point of solidi-
fication is about the same over the entire film cross section. A graded pore dis-
tribution is obtained when the time between precipitation and solidification of
the polymer increases with increasing distance from the skin.
_- --
- -..
- ------ precipitant-
1skin
1skin
r- --jiTcij3tant
_---- -
-- - --
4
jpolymer solution J
+skin
Figure 1.24: Schematic diagram of finger formation at various times during

precipitation.
Cellular and Modular Membrane Fine Structure. Several authors have com-
mented on the two different structures shown in Figure 1.15. It is generally as-
sumed that a fine structure is already predetermined in the casting solution
prior to the actual phase separation.44r4s A cellular structure will be obtained
when in the beginning state of the phase separation, i.e., before the solidification
of the polymer-rich phase, the continuous phase is formed by the polymer-rich
phase. A nodular structure is obtained when the precipitant-rich phase is con-
tinuous, completely surrounding droplets of the polymer-rich phase.
The Effect of the Solvent-Precipitant System on Membrane Structure and
Properties. The precipitant and the solvent used in membrane preparation de-
termine both the activity coefficient of the polymer in the solvent-precipitant
mixture and the concentration of polymer at the point of precipitation and so-
lidification.
The effect of additives to the casting solution or to the precipitant on the

membrane structure can be explained by changes of the activity coefficients of
the polymer, the solvent or the precipitant and with that the rate of precipita-
tion, The effect of additives such as salts, organic solvents, or other “pore-
farmers” to the casting solution or precipitation bath is described in detail
in the literature.36p47
The Effect of the Polymer Concentration in the Casting Solution on the
Membrane Structure. A low polymer concentration in the casting solution tends
to induce precipitation in a finger-type structure, while high polymer concen-
trations tend to induce precipitation of sponge-structured membranes. The ef-
fect of polymer concentration on membrane structures can be explained by the
initiation and propagation of fingers. Higher polymer concentration in the cast-
ing solution produces a higher polymer concentration at the point of precipita-
tion, which will thus tend to increase the strength of the surface of polymer
film first precipitated, and tend to prevent initiation of fingers. Increasing the
viscosity of the casting solution has the same effect.
The Effect of Pre- and Post-Precipitation Procedures on Membrane Struc-
tures and Properties. In the original recipes for making phase inversion mem-
branes, the evaporation step prior to the precipitation of the film was considered
as being essential for the formation of the asymmetric membrane.46 Meanwhile,
it has become apparent that skin-type asymmetric membranes can be prepared
without an evaporation step. The effect of the evaporation results in an in-
crease of the polymer concentration at the surface of the cast film, due to a
loss of solvent. This does not necessarily lead to denser membranes. This is
shown in Figure 1.25, where the salt rejection of cellulose acetate membranes
prepared from a casting solution containing 25% cellulose-acetate, 45% acetone,
and 30% formamide, which was exposed to air at room temperature prior to pre-
cipitation in water at O’C and annealing at 75°C for 2 minutes is shown as a func-
tion of the evaporation time. The filtration tests were carried out at a hydro-
static pressure of 100 bar with a 1% NaCl solution. At short evaporation times
(<I minute), there is no change in rejection, at longer evaporation times (>2
minutes), there is a drastic loss in rejection. This is due to a significant change in
the casting solution composition at the membrane surface because of differ-
ences in the boiling temperatures of acetone (kp = 6O’C) and formamide (kp =
12O’C). During the air exposure time, acetone will preferentially be evapo-
rated because it is the more volatile solvent, while the formamide concentration
will increase in the film surface. Thus, the original acetone-formamide ratio of
the solvent system will be changed. In an independent set of experiments, it was
shown, however, that cellulose acetate membranes made from casting solutions
with formamide concentrations exceeding 35% did not reject NaCl.35
As indicated earlier, membranes made from a casting solution containing
25% cellulose acetate, 45% acetone and 30% formamide do not show any NaCI-
rejection unless they are annealed hot water. Thus, for this type of membrane
the annealing is an essential postprecipitation membrane treatment step. The
effect of the annealing procedure on membrane flux and sale rejection is dem-
onstrated in Figure 1.26 where the flux and salt rejections of membranes pre-
cipitated in ice water from a casting solution of 25% cellulose acetate, 45%
acetone and 30% formamide are shown as a function of the annealing temper-
ature. The annealing time was kept constant at 2 minutes and the membranes
were tested at 100 bars with a 1% NaCI-solution.
I I I I I
1 2 3 4 5
Evaporation time (min)
Figure 1.25: Rejection of a cellulose acetate membrane prepared from a casting
solution containing 25% cellulose acetate, 45% acetone and 30% formamide by
precipitation in a water bath at 0°C as a function of the evaporation time prior
to the precipitation. (Test condition: 1% NaCI-solution, 100 bar hydrostatic
pressure).
-1 100 s3-G
-
C
0
c
v
al
2
L
50 i
I I \ ,
I I
20 40 60 80 100
Annealing temperature (OC)

Figure 1.26: Transmembrane flux and salt rejection of a membrane prepared
from a solution of 25% cellulose acetate, 45% acetone and 30% formamide by
precipitation in water as a function of the past precipitation annealing tempera-
ture. (Test conditions: 1% NaCI-solution, 100 bar hydrostatic pressure).
Homogeneous Membranes
A homogeneous membrane is merely a dense film through which a mixture
of chemical species is transported under the driving force of a pressure, concen-
tration, or electrical potential gradient. The separation of various components
in a solution is directly related to their transport rates within the membrane
phase, which is determined mainly by their diffusivity and concentration in the
membrane matrix.4*42 An important property of homogeneous membranes is
that chemical species of similar size, and hence similar diffusivities, may be sep-
arated when their concentration, that is their solubility in the film, differs sig-
nigicantly. The membrane phase itself may be solid or liquid. The mass trans-
port in homogeneous membranes occurs strictly by diffusion; thus permeabil-
ities are rather low. Homogeneous membranes should, therefore, be as thin as
possible.
Homogeneous Polymer Membranes. Although there are a number of homo-
geneous membranes made from inorganic materials, such as glass or certain met-
als,53r54,the technically more important structures are of polymeric origin. Mod-
ern polymer chemistry is highly proficient in tailoring polymers to specific aims
in terms of mechanical or thermal stability and chemical compatibility. In gen-
eral, mass transfer will be greater in amorphous polymers than in highly crys-
talline or cross-linked polymers.55 Thus, crystallization and orientation are to be
avoided as much as possible when high permeabilities and transmembrane fluxes
are desired. However, physical properties and, in particular, the mechanical
strength of the polymer as well as its selectivity may then be adversely affected,
and the final product will represent a compromise between necessary strength,
selectivity and mass-transfer rates. s6 The principle aim is to create as thin a bar-
rier as possible, consistent with the required strength and absence of pinholes
and defects. The two basic membrane configurations are flat sheets and hollow
fibers.5~5*.Flat sheets can be prepared by casting from solution, by extruding
from a polymer melt or by blow and press molding. Hollow fibers are generally
made by extrusion with central gas injection.59 Because of their high selectivity
for different chemical components homogeneous membranes are used in various
applications, which in general involve the separation of different low molecular
weight components with identical or nearly identical molecular dimensions. The
most important applications of homogeneous polymer membranes are in gas
separation, pervaporation, and reverse osmosis. For the separation of gases sili-
con rubber, because of its relatively high permeability, is the more widely used
basic material.60r61 For reverse osmosis, cellulose esters and various polyamides
serve as the barrier polymer for the membrane preparation.62r63
Most technically utilized homogeneous polymer membranes consist of a
composite structure where a very thin homogeneous selective polymer film is
supported by a thicker microporous structure providing the mechanical strength.
Homogeneous Metal and Glass Membranes. There is only one type of ho-
mogeneous metal membrane of technical importance. This is the palladium, or
palladium-alloy membrane used for the separation and purification of hydrogen.
The permeability of hydrogen in palladium, palladium alloys and several other
metals such as platinum, silver, iron, nickel, etc. is several orders of magnitude
higher than of any other gases.g4 The permeability of hydrogen in palladium al-
loy membranes is highly temperature dependent. The separation is, therefore,
carried out at elevated temperature (m4000C).64 The membranes generally con-
sist of 10 to 50 pm thick metal foils. Because of their high selectivity, these
membranes are used for production of high purity hydrogen (>99.99% Hz). Al-
though the process seems technically feasible, there are only very few commer-
cial plants in operation.65 The same is true for the use of homogeneous silica
glass membranes, the only other homogeneous inorganic material which shows
any promise to be used as selective barrier especially for the separation of he-
lium. Like metal membranes, glass membranes are operated at elevated tempera-
ture. Until today, however, no commercial industrial size plants are in operation.
Homogeneous glass membranes also have a high selectivity for H+-ions, thus they
are used as the selective barrier in pH-electrodes. Their preparation is described
in some detail in the literature and shall not further be discussed in this chapter.
Liquid Membranes. Liquid membranes have gained increasing significance
in recent years in combination with the so-called facilitated transport which uti-
lizes selective “carriers” transporting certain components such as metal-ions se-
lectively and at a relatively high rate across the liquid membrane interphase.66t67
It is relatively easy to form a thin fluid film. It is difficult, however, to maintain
and control this film and its properties during a mass separation process. In order
to avoid a break-up of the film, some type of reinforcement is necessary to sup-
port such a weak membrane structure. Two different techniques are used today
for the preparation of liquid membranes. In the first case, the selective liquid
barrier material is stabilized as a thin film by a surfactant in an emulsion-type
mixture.6ar6g .In the second technique for making liquid membranes, a micro-
porous polymer structure is filled with the liquid membrane phase.mt71.1n this
configuration, the microporous structure provides the mechanical strength and
the liquid-filled pores the selective separation barrier. Both types of membranes
are used today on a pilot-plant stage for the selective removal of heavy metal-
ions or certain organic solvents from industrial waste streams. They have also
been used rather effectively for the separation of oxygen and nitrogen.n
Supported Liquid Membranes The preparation of supported liquid mem-
branes is extremely simple, when certain requirements concerning the selective
barrier and the microporous support material are fulfilled. The liquid membrane
material should have a low viscosity and low vapor pressure, i.e., high boiling
point and, when used in aqueous solutions, a low water solubility. Otherwise,
the useful lifetime of the membrane is rather limited. The microporous substruc-
ture should have a high porosity, a pore size small enough to support the liquid
membrane phase sufficiently under hydrostatic pressure and the polymer of the
substructure should be hydrophobic in nature for most liquid membranes used
in contact with aqueous feed solutions. In practice, liquid membranes are pre-
pared by soaking a hydrophobic microporous membrane, such as a Goretex
(Gore Corp.) or Cellgard (Celanese Corp.) type stretched polytetrafluorethylene
or polyethylene membrane, in the hydrophobic liquid which may consist of
a selective carrier such as certain oximes or tertiary or quaternary amines dis-
solved in kerosene. The disadvantage of supported membranes is their thickness
which is determined by the thickness of the microporous support structure,
which is in the range of IO to 50 pm, and therefore about 100 times the thick-
ness of the selective barrier of an asymmetric polymer membrane. Thus, the
fluxes of supported liquid membranes can be low even when their permeabilities
are high.n
Unsupported Liquid Membranes. Very thin unsupported liquid membranes
may be obtained, when the selective membrane material is stabilized by an aPPro_
priate surfactant in an aqueous emulsion. The preparation procedure is indicated

in Figure 1.27. A hydrophobic membrane phase is transformed into an emulsion
by stirring with an aqueous phase. Ideally, droplets form in this process, in which
the aqueous phase is surrounded by a relatively thin hydrophobic membrane
forming phase which is surrounded by a second aqueous phase. The mass ex-
change occurs between the inner and outer aqueous phases through the liquid
membrane interphase. In reality, the hydrophobic membrane phase and the
surrounding aqueous phases are more fractionated and the diffusion pathways
become longer as a result. With another aqueous solution, the component to
be eliminated is supplied to the original emulsion and passes through the mem-
brane into the internal solution.
stripping agent
liquid membrane
mode, trtatt’“tnt practical application
\- liquid mcmbta-
Ccparation mechanism in unsupported liquid mrmbranrs
Figure 1.27: Schematic diagram showing the formation of an emulsified liquid

membrane.
Ion-Exchange Membranes. Ion-exchange membranes consist of highly

swollen gels carrying fixed positive or negative charges. The properties and
preparation procedures of ion exchange membranes are closely related to those
of ion-exchange resins.74 As with resins, there are many possible types with
different polymer matrixes and different functional groups to confer ion-ex-
change properties on the product. Although there are a number of inorganic
ion-exchange materials, most of them based on zeolites and bentonites, these
materials are rather unimportant in ion-exchange membranes and will not be
discussed further.
There are two different types of ion-exchange membranes: (1) cation-ex-

change membranes which contain negatively charged groups fixed to the poly-
mer matrix, and (2) anion-exchange membranes which contain positively charged
groups fixed to the polymer matrix. In a cation-exchange membrane, the fixed
anions are in electrical equilibrium with mobile cations in the interstices of the
polymer, as indicated in Figure 1.28. This figure shows schematically the ma-
trix of a cation-exchange membrane with fixed anions and mobile cations, which
are referred to as counter-ions. In contrast, the mobile anions, called co-ions, are
more or less completely excluded from the polymer matrix because of their elec-
trical charge which is identical to that of the fixed ions. Due to the exclusion of
the co-ions, a cation-exchange membrane permits transfer of cations only. An-
ion-exchange membranes carry positive charges fixed on the polymer matrix.
Therefore, they exclude all cations and are permeable to anions only. The most
required properties from ion-exchange membranes are:
l High permselectivity-an ion-exchange membrane should be highly

permeable for counter-ions, but should be impermeable to co-ions.
l Low electrical resistance-the permeability of an ion-exchange mem-
brane for the counter-ions under the driving force of an electrical
potential gradient should be as high as possible.
0 Good mechanical and form stability-the membrane should be me-
chanically strong and should have a low degreeof swellingor shrink-
ing in transition from dilute to concentrated ionic solutions.
l High chemical stability-the membranes should be stable over a
pH-range from 1 to 14 and in the presence of oxidizing agents.
&Matrix with Fixed Charges
OC ounter- Ion
0 Co-Ion
Figure 1.28: Schematic diagram of the structure of a cation-exchange membrane

showing the polymer matrix with the negative fixed charges, the positive counter-
ions, and the negative co-ions.
It is often difficult to optimize the properties of ion-exchange membranes

because the parameters determining the different properties often act contrary.
For instance, a high degree of crosslinking improves the mechanical strength of
the membrane but also increases its electrical resistance. A high concentration of
fixed ionic charges in the membrane matrix leads to a low electric resistance but,
in general, causes a high degree of swelling combined with poor mechanical sta-
bility. The properties of ion-exchange membranes are determined by two para-
meters, i.e., the basic polymer matrix and the type and concentration of the
fixed ionic moiety. The basic polymer matrix determines to a large extent the
mechanical, chemical and thermal stability of the membrane. Very often the ma-
trix of an ion-exchange membrane consists of hydrophobic polymers such as
polystyrene, polyethylene or polysulfone.75 Although these basic polymers are
insoluble in water and show a low degree of swelling, they may become water
soluble by the introduction of the ionic moieties. Therefore, the polymer ma-
trix of ion-exchange membranes is very often crosslinked. The degree of cross-
linking then determines to a large extent the degree of swelling, and the chem-
ical and thermal stability, but also has a large effect on the electric resistance and
the permselectivity of the membrane.76r77.
The type and the concentration of the fixed ionic charges determine the
permselectivity and the electrical resistance of the membrane, but they also have
a significant effect on the mechanical properties of the membrane. The degree of
swelling, especially, is affected by the fixed charge concentration. The follow-
ing moieties are used as fixed charges in cation-exchange membranes:78-84
-SOB- , -coo- , -Po32- , -AsOj2-
In anion-exchange membranes, fixed charges may be:80,*l
-NHJ+ , :NH2+ , $ N+ , 5s’.
These differently charged groups have a significant effect on the ion-exchange

behavior of the membrane. The sulfonic acid groups, e.g., -SO; is completely
dissociated over nearly the entire pH-range, while the carboxylic acid group
-COO-is virtually undissociated in the pH-range <7. The quaternary ammonium
group again is completely dissociated over the entire pH-range, while the primary
ammonium group is only weakly dissociated. Accordingly, ion-exchange mem-
branes are referred to as being weak or strong acid or basic in character.s2 Most
commercial ion-exchange membranes can be divided, according to their struc-
ture and preparation procedure, into two major categories, i.e., homogeneous
and heterogeneous membranes. In general, heterogeneous ion-exchange mem-
branes have several disadvantages, the most important of which are relatively
high electrical resistance and poor mechanical strength when highly swollen in
dilute salt solutions.
Homogeneous ion-exchange membranes have significantly better properties
in this respect, since the fixed ion charges are distributed homogeneously over
the entire matrix.
Preparation Procedure of Homogeneous Ion-Exchange Membranes. The
methods of making homogeneous ion-exchange membranes can be summarized
by three different basic procedures:
(I) Polymerization or polycondensation of monomers, of which at least one

must contain a moiety that either is or can be made anionic or cationic.
The first membranes made by this procedure were prepared from phenol by
polycondensation with formaldehyde according to the following reaction
scheme:‘*
&%&...-&~2-~r~~-
..
803-H+ S 03. H+ 3-
Phenol is treated with concentrated Hz!504 at elevated temperatures leading

to phenolsulfonic acid in paraform, a brown crystalline material. The phenolsul-
fonic acid and a solution of formaldehyde in water is treated at elevated temper-
atures for several hours. The solution is cast into a film. Excess monomers are re-
moved by washing the film in water.
A very common method of preparing a cation- or anion-exchange membrane
is the polymerization of styrene and divinylbenzene and subsequent sulfonation
or amination. The cation-exchange membrane is obtained by the following re-
action scheme:83
HC=CH2
. ..-CC-CHz-... . ..-CH-CHz -...
The anion-exchange group is introduced into the polymer by chloromethyiation

and amination with trimethylamine according to the following reaction scheme:
There are numerous literature references to the preparation of ion-exchange

membranes by polymerization.
(2) Introduction of anionic or cationic moieties into a preformed film by
techniques such as imbibing styrene into polymer films, polymerizing the im-
bibed monomer, and then sulfonating the styrene.
Starting with a film makes the membrane preparation rather easy. The start-
ing material may be a film from a hydrophilic polymer, such as cellophane or
polyvinyl alcohol, or a film from a hydrophobic polymer, such as polyethylene
or polystyrene. ion-exchange membranes made by sulfochlorination and amina-

tion of polyethylene sheets have low electrical resistance combined with high
permselectivity and excellent mechanical strength. The reaction scheme for prep-
aration of ion-exchange membranes by sulfochlorination and amination of poly-
ethylene sheets is given below.”
Sulfochlorination
-CH~CH;cHj + SO,*CI, - FH -CH,-CH_j l HCI
SW
Hydrolysis
-fH-CH,-CH; + sNa OH __, -CH-C%CHi + NaCl + H,o
SC&Cl &O;Na+
Amination
-$H-CH2-CY - + I$ N+-CH, W -7 -Cl+ - + HCI
sop R
So2 -NH-&H3
R
Quaternization Anion-exchanog memhmnp.
+Cl$Br __, -CH-CH2-CH3 -

-
-SrH-CH2-CH2 I
SO2 -NH-$+CH, I FH3
SO2 -NH-+CH3 Bi
R
R
(3) Introduction of anionic or cationic moieties into a polymer chain such

as polysulfone, followed by dissolving the polymer and casting it into a film.
Membranes made by any of the above methods may be cast around screens
or other reinforcing materials to improve their strength and dimensional stabil-
ity. The reaction scheme for the sulfonation of polysulfone is as follows:*5
[-Q- o-~-~;~-O-o- SO*-] +S03 +NaOH
-[-Q-o-~-~~-o-~-302-]
n
B'OiNa+
The sulfonated polysulfone can be cast into a film on a screen and precipitated
after most of the solvent has been evaporated, leading to a reinforced membrane
with excellent chemical and mechanical stability and good electrochemical prop-
erties.
Heterogeneous Membranes. These membranes consist of fine colloidal ion-
exchange particles embedded in an inert binder such as polyethylene, phenolic
resins or polyvinyl chloride.66 Such a membrane can be prepared simply by cal-
endering ion-exchange particles into an inert plastic film. Another procedure is
dry molding of inert film-forming polymers and ion-exchange particles and then
milling the mold stock. Also, ion-exchange particles can be dispersed in a solu-
tion containing a film-forming binder, and the solvent evaporated to give an ion-
exchange membrane. Similarly, ion-exchange particles can be dispersed in a par-
tially polymerized binder polymer, and the polymerization subsequently com-
pleted. Heterogeneous membranes with usefully low electrical resistances con-
tain more than 65% by weight of the crosslinked ion-exchange particles. Since
these ion-exchange particles swell when immersed in water, it has been difficult
to achieve adequate mechanical strength combined with low electrical resistance.
Most heterogeneous membranes that possessadequate mechanical strength gen-
erally show poor electrochemical properties. On the other hand, a membrane
that contains ion-exchange particles large enough to show an adequate electro-
chemical performance exhibits poor mechanical strength.
Special Property Membranes. In the literature, there are numerous methods
reported for the preparation of ion-exchange membranes with special proper-
ties s’-~ for instance, for use as battery separators, ion-selective electrodes, or in
the’chlor-alkali process. Especially membranes recently developed for the chlor-
alkali industry are of commercial significance. These membranes are based on
polytetrafluoroethylene and carry sulfone groups in the bulk of the membrane
phase and carboxyl-groups on the surface as the charged moiety. They combine
good chemical stability with high selectivity and low electric resistance.
The structure of a membrane produced by Asahi Chemical is shown in the
following diagram:W
r CF?-CF2), - r CF,-CF I,,, -_( CF,-;F 1”

I! 0
& !F
I * iz
cFJ:F CFJFF
0
? I l/tm t n) 1 6 - 8
as-40
( y24
m/n
( fVq
al-4
SO,Na COONa "'
Significant effort has also been concentrated on the development of anion-

exchange membranes with low fouling tendencies. In a conventional electro-
dialysis plant, the limiting current density through the anion-exchange mem-
brane is lower than in the cation-exchange membrane, largely due to the risk of
precipitation of inorganic or organic negatively charged materials. As a conse-
quence, its electrical resistance may rise during operation. To overcome this
problem, different companies produce special anionic membranes, which are
characterized by the fact that they can be used at a higher current density. In
general, the permselectivity of these membranes is lower than of regular mem-

branes.
Composite Membranes
In processes such as reverse osmosis, gas separation and pervaporation, the
actual mass separation is achieved by a solution-diffusion mechanism in a homo-
geneous polymer layer. Since the diffusion process in a homogeneous polymer
matrix is relatively slow, these membranes should be as thin as possible, as in-
dicated before. Therefore, an asymmetric membrane structure is mandatory for
these processes. Unfortunately, many polymers with satisfactory selectivities and
permeabilities for the various components in gas mixtures or liquid solutions are
not well suited for the preparation of asymmetric membranes by the phase in-
version process described before. This has led to the development of the so-
called composite membranes. A composite membrane is shown schematically in
Figure 1.29. It is composed of a 20 to 100 nm thin dense polymer barrier layer
formed over an approximately 100 I.tm thick microporous film. Composite mem-
branes differ from asymmetric reverse osmosis membranes by the mode of prep-
aration which consists of two steps: (I) casting of the microporous support, and
(2) deposition of the barrier layer on the surface of this microporous support
layer. This preparation mode leads to significant advantages of the composite
membrane compared to the integral asymmetric membrane. In an integral asym-
metric membrane, the selective barrier layer and the microporous support al-
ways consist of the same polymer. In a composite membrane, different poly-
mers may be-and in general are-used for the microporous support and the selec-
tive barrier layer. This means polymers which show the desired selectivity for a
certain separation problem, but have poor mechanical strength or poor film-
forming properties, and which are therefore not suited for preparation into in-
tegral asymmetric membranes, may well be utilized as the selective barrier in
composite membranes. This, of course, expands the variety of available materi-
als for the preparation of semipermeable membranes.
/Porous Support
Figure 1.29: Schematic diagram of an asymmetric composite membrane show-

ing the microporous support structure and the selective skin layer.
Preparation Procedures of Composite Membranes. In making composite

membranes, two completely different tasks have to be solved. One is the prepar-
ation of a suitable microporous support and the second task is the preparation of
the actual barrier layer and laminating it to the surface of the support film. The
performance of a composite membrane is not only determined by the proper-
ties of the selective barrier layer, but it is also significantly effected by properties
of the microporous support film.
preparation and Deposition of the Selective Barrier Layer on the Micro-
porous Support. The techniques used for the preparation of composite struc-
tures may be grouped into four general types:91+5
(I) Casting of the barrier layer separately, e.g., on the surface of a wa-
ter bath followed by lamination to the microporous support film.
(2) Dip-coating of the microporous support film in a polymer, a
reactive monomer or prepolymer solution followed by drying
or curing with heat or radiation.
(3) Gas-phase deposition of the barrier layer of the microporous sup-
port film from a glow discharge plasma.
(4) Interfacial polymerization of reactive monomers on the surface of
the microporous support film.
Casting an ultrathin film of cellulose acetate on a water surface and trans-

ferring the film on a microporous support was one of the earliest techniques
used for preparing composite reverse osmosis membranes for water desalination.
The actual selective barrier was prepared by dissolving 2-10 percent cellulose di-
acetate in a solvent exhibiting slight water solubility such as cyclohexanone.
Casting an ultrathin film from a dilute acetone solution on a glass plate and re-
leasing the film from the plate after the evaporation of the acetone by immer-
sion in water was another method of preparing ultrathin selective barriers. Al-
though both preparation techniques lead to barrier layers of less than 100 nm
with correspondingly high flux rates, they are not well suited for large scale in-
dustrial production.%
Dip coating a microporous support membrane in polymer or prepolymer
solution was also first developed for the preparation of reverse osmosis mem-
branes. Here, a microporous membrane prepared from mixed cellulose esters was
first coated by a protective layer of polyacrylic acid to prevent the solvent of the
casting solution of the barrier layer, which consisted, e.g., of cellulose triacetate
in chloroform, from dissolving the support membrane. This technique was later
improved by using a microporous sublayer, which had better overall mechanical
and thermal stability and which was insoluble in the solvent of the barrier layer,
such as an “open” polysulfone ultrafiltration membrane. Today, dip coating is
applied mainly for the preparation of composite membranes to be used in gas
separation and pervaporation.97 Particularly, polymers such as polydimethyl-
siloxane, which are available as still soluble prepolymers that can easily be cross-
linked by a heat treatment procedure thus becoming insoluble in most solvents,
are suited for the preparation of this type of composite membrane. If the pore
dimension in the support membrane is selected properly, the prepolymer is un-
able to penetrate the support and a rather thin uniform barrier layer of 0.05 to
1 pm thickness can easily be prepared. A typical composite membrane prepared
by dip coating an asymmetric polysulfone ultrafiltration membrane into a 1
wt % solution of polydimethylsiloxane followed by thermal crosslinking is
shown in the scanning electron micrograph of Figure 1.30.
Figure 1.30: Scanning electron micrograph showing a composite membrane

with polydimethylsiloxane as the selective layer deposited on a polysulfone sup-
port membrane.
Gas phase deposition of the barrier layer on a dry microporous support

membrane by plasma polymerization was also successfully used for the prepara-
tion of reverse osmosis membranes. Many organic compounds having adequate
vapor pressure can be used to form a barrier layer on a microporous support.
The plasma reactions are rather heterogeneous not only involving polymeriza-
tion but depolymerization and modification of functional groups. Although re-
verse osmosis membranes with excellent desalination properties showing salt re-
jection in excess of 99%, and fluxes of 1.5 m3mzd-’ when tested with seawater
have been prepared on a laboratory scale, 98 large scale industrial production util-
izing plasma polymerization for the preparation of composite membranes seems
to be difficult.
Today, by far the most important technique for preparing composite mem-
branes is the interfacial polymerization of reactive monomers on the surface of a
microporous support film. The first membrane produced on a large scale with
excellent reverse osmosis desalination properties was developed in the early sev-
enties in the North Star Research Institute under the code name NS 100.” The
preparation procedure of this membrane, which exhibited water fluxes of about
1 m3m3d-’ and salt rejections in excess of 99% when tested with seawater at 60
bar pressure, was rather simple. A polysulfone support membrane was soaked in
an aqueous solution of 0.5 to 1% polyethyleneimine, which was reacted inter-
facially at the membrane surface with a 0.2 to 1% solution of toluene diiso-
cyanate in hexane. A heat curing step at 110°C leads to further crosslinking
of the polyethyleneimine. The preparation is shown schematically in Figure
1.31. The process seems to involve two types of reactions. In a first step, the
polyethyleneimine reacts rapidly at the interphase with the toluene diisocyanate
to form a polyamide surface skin while amine groups below this surface remain
unreacted. In the second heat treatment step, internal crosslinking of the poly-
ethyleneimine takes place. Thus, the final membrane has three distinct layers of
increasing porosity: (I) the dense polyamide surface skin which acts as the ac-
tual selective barrier, (2) a thin crosslinked polyethyleneimine layer which ex-
tends into the pore of the support film, and (3) the actual polysulfone support
membrane.
PEI +
Figure 1.31: Schematic diagram showing the formation of a composite mem-

brane by interfacial polymerization of polyethyleneimine with toluene diisocy-
anate.
Although the NS-100 membrane showed significantly higher salt rejection

capability and higher fluxes than most integral asymmetric reverse osmosis mem-
branes, further improvements were achieved by using aromatic diamines and tri-
acyl chloride reactants. One of these membranes, the FT-30, produced by Film
Tee Corporation from monomeric diamine reactants such as m-phenylenedi-
amine interfacially polymerized with trimesoyl chloride shows not only excel-
lent desalination properties but also highly improved stability towards oxidizing
agents.ree
Preparation of the Microporous Support of Composite Membranes. Al though
the selectivity of a composite membrane is determined nearly exclusively by the
actual barrier layer, its overall performance is strongly affected by the micro-
porous substructure. The mechanical stability of the membrane as well as its
flux rate is to a large extent determined by the pore size and the overall porosity
of the substructure as can easily be demonstrated by the schematic diagram of
Figure 1.32. This diagram shows an idealized homogeneous barrier layer with a
thickness of 0.1 pm on a microporous support, with the average pore diameter

being 0.2 pm and a porosity of 50%. The actual diffusion pathway of thecompo-
nent through the barrier layer is always longer than the thickness of the layer.
The average path length, hM, can, to a first approximation, be expressed as a
function of the membrane overall porosity, E, the thickness of the barrier layer,
ho, and the radius of the pores, r, in the support structure, by the following
relation:
= E X0 + (l--E) (12)
L.M
This relation which can be obtained by simple geometric considerations in-

dicates that the effective diffusional path length is always longer than the thick-
ness of the barrier layer and that it is strongly affected by the overall porosity of
the substructure. For practical purposes, the porosity of the support should be
as high as possible. Unfortunately, many ultrafiltration membranes have a rather
low surface porosity of 24%. While at relatively high surface porosities, i.e., in
excess of 50%, the effective diffusional length is approximately identical with
the thickness of the barrier layer it will increase according to equation (12) by
about an order of magnitude, when the porosity is less than 1% assuming the
barrier layer thickness is approximately the same as the pore radius. Accord-
ingly, the flux will decrease by the same magnitude. The geometrical considera-
tion expressed in equation (12) indicates that the surface porosity of the support
layer will significantly effect the membrane flux, and it should always be as high
as possible when optimal flux rates shall be achieved with thin film composite
membranes. For mechanical strength, the pore diameter should not be signifi-
cantly larger than the film thickness.
Figure 1.32: Schematic diagram showing a homogeneous barrier film on a mi-

croporous substructure. The actual average diffusional pathlength hi of a com-
ponent through a film of the thickness ho is expressed as a function of the over-
all porosity and the pore radius.
Industrial Scale Membrane Production

For any membrane to be useful in an industrial process it has to be produced
on a large scale and installed into an appropriate device which should be com-
pact, reliable, and inexpensive. For technical use membranes are, therefore, in-
tegrated into so-called modules. Besides economic considerations, chemical en-
gineering aspects are of prime importance for the design of membrane modules.
Since the various membrane separation processes differ significantly in their op-
erational concept and their applications, the membrane modules used for the
various processes are equally different.
Membrane Modules and Their Fabrication. A large number of different
module systems are described in the literature. However, only six basic types are
used today on a large industrial scale. These modules are shown schematically in
Figure 1.33 (a)-(f). The pleated cartridge filter module, which is shown in (a) is
used mainly in dead-end microfiltration. It consists of a pleated membrane car-
tridge installed in a pressurized housing. The feed solution enters the filter from
the housing side and the product is collected in a center tube which is sealed
against the housing by an O-ring.
Cartridge type filters are operated at relatively low hydrostatic pressures.
Their useful life is limited due to plugging of the membrane pores by retained
solutes. The actual cartridge is made by pleating a membrane sheet and potting
the ends by an appropriate resin or hot-melt-glue as indicated in Figure 1.33 (a).
Another module type used on an industrial scale for various membrane sepa-
ration processes including ultrafiltration, reverse osmosis, and gas separation is
the plate-and-frame module. Its design has its origins in the conventional filter
press concept. The membranes, porous membrane support plates, and spacers
forming the feed flow channel are clamped together and stacked between two end-
plates as indicated in the schematic diagram of Figure 1.33 (b). There are various
types of plate-and-frame modules on the market which offer, however, only
slight variations in their basic configuration.“’
A variation of the basic plate-and-frame concept is the spiral-wound module,
which is widely used today in reverse osmosis, ultrafiltration, and gas separation.
Its basic design is illustrated in Figure 1.33 (c). The feed flow channel spacer, the
membrane, and the porous membrane support are rolled up and inserted into an
outer tubular pressure shell. The filtrate is collected in a tube in the center of the
roll.
While the previously described three membrane modules required flat sheet
membrane material for their preparation, special membrane configurations are
needed for the preparation of the tubular, capillary, and hollow fiber modules.
The tubular membrane module consists of membrane tubes placed into porous
stainless steel or fiber glass reinforced plastic pipes. The pressurized feed solution
flows down the tube bore and the permeate is collected on the outer side of the
porous support pipe, as indicated in Figure 1.33 (d). The diameters of tubular
membranes are typically between l-2.5 cm. In some modules, the membranes
are cast directly on the porous pipes and in others they are prepared separately
as tubes and then installed into the support pipes.
The capillary membrane module, which is shown schematically in Figure
1.33 (e), consists of a large number of membrane capillaries with an inner di-
ameter of 0.2 to 3 mm arranged in parallel as a bundle in a shell tube. The feed
solution is passed down the center of the membrane capillary and the filtrate,
which permeates the capillary wall, is collected in the shell tube. The capillary
membrane module requires as basic material membranes in a self-supporting ca-
pillary configuration, which, when asymmetrically structured, carry the selective
--
/
b)
(4
Figure 1.33: Schematic diagram showing membrane modules presently used in

industrial separation processes: (a) pleated membrane filter cartridge; (b) plate-
and-frame membrane module; (c) spiral wound membrane module; (d) tubular
membrane module; (e) capillary membrane module; (f) hollow fiber membrane
module.
Feed solution
=zzzEE?
Feed solution Capillary membrane Shell tube Concentrate

/ I I 1 I
1 Filtrate
\E Pory realnr
Figure 1.33: (continued)

barrier on the inner side of the capillary as indicated in the scanning electron
micrograph of Figure 1.34, which shows a typical capillary ultrafiltration mem-
brane prepared in a wet-spinning process.
20 IJm
~ ~
Figure 1.34: SEM of a capillary membrane.
The same basic spinning process is used for the preparation of hollow fiber
membranes, which have an outer diameter of 50 to 100 ,urn. In hollow fiber
membranes, the selective layer is on the outside of the fibers, which are installed
as a bundle of several thousand fibers in a half loop with the free ends potted
with an epoxy resin in a pressure tube as indicated in Figure 1.33 If). The feed
solution is introduced around th'e outside of the hollow fibers. The filtrate passes
through the fiber walls and flows up the bore to the open end of the fibers at the
epoxy head.
Membrane Manufacturing Equipment

Based on the different modules used in technical scale membrane separation
processes, there are four basic membrane configurations produced today on a
large scale. These are flat sheet, tubular, capillary, and hollow fiber membranes.
Flat sheet membranes are generally manufactured on a casting machine,
which is shown schematically in Figure 1.35. A polymer solution is cast by a
casting knife on a polyester or polyethylene support paper, which is continu-
ously supplied from a roll. The cast polymer film is fed to the precipitation bath,
where the actual membrane is formed. After a certain residence time in a rinse
bath, where residue solvent is removed, the membrane is collected as a flat sheet
on the take up roll. The membranes, which are obtained as up to 2 111wide con-
tinuous sheets, are then further processed into the desired module configuration.
Tensioning Roller
Solution Trough
Doctor
\ r-7 abric Roll
/Spreader Roller
Tap Water
Figure 1.36: Schematic diagram indicating the function of a casting machine

used for the preparation of supported flat sheet membranes.
Capillary and hollow fiber membranes are manufactured generally by a wet-

spinning process using a spinneret with a double bore nozzle as indicated in Fig-
ure 1.36 (a), which shows a schematic diagram of a spinneret for the production
of asymmetric hollow fiber or capillary membranes. The casting solution is al-
ways fed in the outer bore of the nozzle. If an asymmetric capillary membrane
with the skin on the inside shall be produced, the precipitant is directed through
the central bore, thus, the precipitation progresses from the inside to the outside
and a capillary with the selective layer on the inside is formed immediately at
the outlet of the spinneret. The capillaries, therefore, have about the dimensions
of the spinning nozzle.
If an asymmetric hollow fiber with the skin on the outside is to be produced,
the precipitant in the inner bore is replaced by an inert gas and the fiber is spun
into the precipitation bath. Between the precipitation bath and the spinneret
there is an air gap as indicated in Figure 1.36 (b) where the fiber may be drawn
to obtain the desired dimensions before precipitation. Hollow fibers have, there-
fore, often significantly smaller diameters than the nozzle.
Tubular membranes are prepared either by an ultrasonic welding process
from flat sheet membranes or by direct casting on a porous support tube using a
conical casting device pulled through the porous tube. Since the outer diameter
of the casting cone is slightly smaller than the inner diameter of the tube, a thin
polymer solution film is formed on the inside of the tube, which is then con-
verted into a membrane by immersing the entire tube into a precipitation bath.
There are many variations of the basic membrane preparation techniques de-
scribed in this chapter. But most of them have only limited technical significance
and shall not be discussed further.
Central fluld
solution
Hollorvnsed*/ \ Tubs-in-orifice
t F- Spinneret
(adjustable)
Fiber take-up 7
+
Precipitation bath
Figure 1.36: Manufacturing device for capillary and hollow fiber membranes:
(a) spinneret; (b) hollow fiber uptake unit.
FUTURE DEVELOPMENTS
For many separation processes, such as sea and brackish water desalination
by electrodialysis and reverse osmosis or the production of ultrapure water by
microfiltration, membranes with quite satisfactory properties are commercially
available today. But, even for comparatively mature processes such as microfil-
tration and reverse osmosis, “better” membranes are desirable. In microfiltra-
tion, research is concentrated on membranes having an asymmetric structure
with high surface porosities in excess of 70% and average pore sizes at the sur-
face between 0.01 and 0.1 E.tm.These membranes are made from polymers, such
as Nylon 6.6 which show good solvent resistance and low adsorption for “foul-
ing” materials. In reverse osmosis water desalination work is concentrated mainly
on the development of membranes with improved stability against oxidizing
agents such as chlorine or hypochlorite with higher fluxes and better flux stabil-
ity. With the emerging use of microfiltration, ultrafiltration, reverse osmosis, and
electrodialysis in biotechnology and the chemical process industry, completely
new requirements are put on the membranes to be used in these applications.
The main effort as far as synthetic membranes are concerned isconcentrated
on the development of completely new membranes for processes such as pervap-
oration, gas separation, membrane distillation, or as ion transferring separators in
batteries, fuel cells or electrochemical production processes. Liquid membranes
with selective carriers used today for the separation and concentration of heavy
metal ions or certain organic compounds are being developed further to be used
in gas separation.
A very important area of today’s ongoing research is the development of
functional synthetic membranes which mimic the function of the biological
membrane. lo2 Biocatalytic membranes, energy and information transducing
membranes have already been produced on a laboratory scalelo and are used to-
day as biosensors for monitoring devices. Although a significant number of func-
tional synthetic membranes have been developed, they are all far behind the ac-
tual biological membranes. But basic research in membrane technology com-
bined with progress in molecular engineering should help to improve the proper-
ties of functional synthetic membranes and thereby increase their use far beyond
today’s level.
REFERENCES
1. Lonsdale, H.K.,J. Memb. Sci., 10 (1982) 81-181.

2. Porter, M.C., Membrane filtration, in: P.A. Schweitzer (Ed.), Handbook of
Separation Techniques for Chemical Engineers, McGraw-Hill, New York
(1979), pp. 2-3 to 2-103.
3. Baker, R. (Ed.), Controlled Release of Bioactive Materials, Academic Press,
New York (1980).
4. Strathmann, H., Trendsin Biotechnology, 3 (1985) 112-118.
5. Strathmann, H., Trennung von molekularen Mischungen mit Hilfe syn-
thetischer Membranen, Steinkopff Verlag, Darmstadt (1979).
6. Katchalsky, A. and Curran, P.T., Nonequilibrium Thermodynamics in Bio-
physics, Harvard University Press, Cambridge, Mass. (1967).
7. Kedem, 0. and Katchalsky, A., Trans. Faraday Sot., 59 (1963) 1918.

8. Helfferich, F., /on-Exchange, McGraw-Hill, London (1962).
9. Hwang, S.-T. and Kammermeyer, K., Membranes in Separations, John Wiley
81 Sons, New York (1975).
10. Druin, M.L., Loft, J.T. and Plovan, S.G., U.S. Pat. 3,801,404 (1974).
11. Gore, R.W., U.S. Pat. 3,962,153 (1976).
12. Schneider, K. and van Gassel, T.J., Chem. Ing. Tech., 7 (1985) 514-521.
13. Fleischer, R. L., Price, P.B. and Walker, R.M., Sci. Amer., 220 (1969) 30.
14. Riedel, C. and Spohr, R., ./. Membr. Sci., 7 (1980) 225-234.
15. Porter, M.C., Am. Lab., November 1974.
16. Reid, H.L., Barnes, A.J., Lock, P.J. and Dormandy, J.A., J. C/in. Path., Vol.
29, No. 9 (September 1976).
17. Kesting, R.E., Synthetic Polymeric Membranes, McGraw-Hill, New York
(1971).
18. Strathmann, H., Production of microporous media by phase inversion proc-
esses, in: Material Science of Synthetic Membranes, D.R. Lloyd (Ed.),
pp. 165-195, ACS Symposium Series 269, American Chemical Society,
Washington D.C. (1985).
19. Zsigmondy, R., U.S. Pat. 1,421,341 (1922).
20. Pall, D.8.. U.S. Pat. 4,340,479 (1982).
21. Hiatt, W.C., Vitzthum, G.H., Wagener, K.B., Gerlach, K. and Josefiak, C.,
Microporous membranes via upper critical temperature phase separation,
in: Material Science of Synthetic Membranes, D.R. Lloyd (Ed.) PP. 229-
244, ACS Symposium Series 269, American Chemical Society, Washing
ton D.C. (1985).
22. MacLean, D.W., J. New England Water Works Assoc., 72 (1958) 272.
23. Cotton, R.A. and Fifield, C.W., Standardization of membrane filters for mi-
crobiological applications, in: B.J. Dutka (Ed.), Membrane Filtration:
Applications, Techniques, and Problems, Marcel Dekker, New York
(1981). pp. 19-39.
24. Hood, H.P. and Nordberg, M.E., U.S. Pat. 2,106,744 (1938).
25. Schnabel, R., Proceed. 5th Intern. Symp., Fresh water from the sea, Vol. 4
(1976) 409-413.
26. Michaels, A.S., Pure Appl. Chem., 46 (1976) 193-204.
27. Kesting, R.E., Phase inversion membranes, in Material Science of Synethetic
Membranes, D.R. Lloyd (Ed.), pp. 131-164, ACS Symposium Series 269,
American Chemical Society, Washington DC. (1955).
28. Loeb, S. and Sourirajan, S., U.S. Patent 3,133,132 (1964).
29. U. Merten (Ed.), Desalination by Reverse Osmosis, MIT Press, Cambridge,
Mass. 11966).
30. Manjikian, S., Ind. Eng. Chem. Prod. Res. Dev., 6 (1967) 23.
31. Zsigmondy, R. and Carius, C., Chem. Ber., 60 B (1927) 1074.
32. Kesting, R.E., J. Appl. Polym. Sci., 17 (1973) 1771.
33. Strathmann, H. and Kock, K., Desalination, 21 (1977) 241.
34. Wijmans, J.G., Baaij, J.P.B. and Smolders, C.A., J. Membr. Sci., 14 (1983)
263-274.
35. Strathmann, H., Scheible, P. and Baker, R.W., J. Appl. Polym. Sci., 15
(1971) 811-828.
36. Strathmann, H., Kock, K., Amar, P. and Baker, R.W., Desalination, 16
(1975) 179-302).
37. Broens, L., Altena, F.W. and Smolders, C.A., Desalination, 32 (1980) 33.
38. Frommer, M.A., Feiner, I., Kedem, 0. and Block, R., Desalination, 7
(1970) 393.
39. Altena, F.W. and Smolders, CA., Macromolecules, 15 (1982) 1491.
40. Haase, R., ThermodynamikderMischphasen,Springer Verlag, Berlin (1956).
41. Jost, W., Diffusion in Solids, Liquids and Gases, Academic Press, New York,
N.Y. (1970).
42. Frommer, M.A., Matz, R. and Rosenthal, U., lnd. Eng. Chem. Prod. Res.
Dev., 10 (1971) 193.
43. Ray, R.J., Krantz, W.B. and Sani, R.L., J. Membr. Sci., 23 (1985) 155-182.
44. Kamide, K. and Manabe, S., Role of microphase separation phenomena in
the formation of porous polymeric membranes, in: Material Science of
Synthetic Membranes, D.R. Lloyd (Ed.), pp. 197-228, ACS Symposium
Series 269, American Chemical Society, Washington D.C. (1985).
45. Wijmans, J.G., Synthetic membranes, Ph.D. Thesis, Twente University,
Enschede, Holland (1984).
46. Sourirajan, S. and Kunst, B., Cellulose acetate and other cellulose ester
membranes, in: Reverse Osmosis and Synthetic Membranes, S. Sourirajan
(Ed.), National Research Council Canada, Ottawa, Canada (1977).
47. Kesting, R.E., Synthetic Polymeric Membranes, A Structural Perspective,
Wiley-Interscience, New York, N.Y. (1985).
48. Lonsdale, H.K., Merten, U. and Riley, R.L., J. Appl. Polym. Sci., 9 (1965)
1341-1362.
49. Stannett, V., Szwarc, M., Ghargava, R.L., Meyer, J.A.. Myers, A.W. and
Rogers, C.E., Permeability of Plastic Films and Coated Paper to Gases
and Vapors, Tappi Monograph Series, No. 23, Technical Association of
the Pulp and Paper Industry, New York (1962).
50. Brubaker, D.W. and Kammermeyer, K., lnd. Engr. Chem., 44 (1952) 1465-
1474.
51. Stern, S.A. and Walawender, W.P., Jr.,Sep. Sci., (1969) 129-159.
52. Stannett, V.T., Koros, W.J., Paul, D.R., Lonsdale, H.K. and Baker, R.W.,
Adv. Polymer Sci., 32 (1979) 99-l 51.
53. Buck, R.P., Thompsen, J.C. and Melroy, O.R., A compilation of ion-selec-
tive membrane electrode literature, in: Ion-Selective Electrodes in Ana-
lytical Chemistry, H. Freiser (Ed.), Plenum Publishing Co., New York
(1978).
54. Kammermeyer, K., Chem. Ing. Tech., 48 (1976) 672-675.
55. Crank, J. and Park, G.S. (Eds.), Diffusion in Polymers, Academic Press,
New York (1968).
56. Stern, S.A., The separation of gases by selective permeation, in: Membrane
Separation Processes, P. Meares (Ed.), Elsevier, Amsterdam (1976).
57. Ward, W.J. Ill, Browall, W.R. and Salemme, R.M.,J. Memb. Sci., 1 (1976)
99-l 08.
58. Richter, J.W. and Hoehn, H.H., U.S. Pat. 3,567,632, March 2, 1971.
59. McLain, E.A. and Mahon, H.I., U.S. Pat. 3,423,491, January 21, 1969.
60. Browall, W.R., Kimura, S.G. and Ward W.J., Ill, Ultrathin membranes for
oxygen enrichment, paper presented at the Interamerican Congress of
Chemical Engineers, Caracas, Venezuela, July 13-l 6, 1975.
61. Product bulletin, OECO Membrane Type Oxygen Enricher, Oxygen En-
richment Co., Schenectady, New York (December 1980).
62. Lonsdale, H.K., Desalination, 13 (1973) 317-332.

63. Lloyd, D.R., Meluch, T.B., Selection and evaluation of membrane mate-
rials for liquid separation, in: Material Science of Synthetic Membranes,
D.R. Lloyd (Ed.), pp. 47-79, ACS Symposium Series 269, American
Chemical Society, Washington, D.C. (1985).
64. Hunter, J.B., U.S. Pat. 2,773,561 (1956).
65. Hunter, J.B., Platinum Metals Rev., 4 (1960) 130.
66. Ward, W.J., I I I, AlChE J., 16 (1970) 405-410.
67. Cussler, E.L.,A/ChE J., 17 (1971) 1300-1303.
68. Li, N.N., U.S. Pat. 3,410,794 (1968).
69. Li, N.N.,A/ChEJ., 17 (1971) 459.
70. Babcock, WC., Baker, R.W., LaChapelle, E.D. and Smith, K.L., J.
Membr. Sci., 7 (1980) 71-87.
71. Largman, T. and Sifniades, S., Hydrometallurgy, 3 (1978) 153-162.
72. Baker, R.W., Roman, J.C., Smith, K.L. and Lonsdale, H.K., industrial
Heating (July 1982).
73. Babcock, WC., Baker, R.W., LaChapell, E.D. and Smith, K.L., J. Memb.
Sci., 7 (1980) 80-100.
74. Kunin, R., Ion-Exchange Resins, Wiley, New York (1958).
75. Bergsma, F. and Kruissink, C.A., Fortschr. Hochpolym.-Forsch.. 2
(1961) 307-362.
76. Meares, P., Trends in ion-exchange membrane science and technology, in:
Ion-Exchange Membranes, D.S. Flett (Ed.), E. Horwood Ltd., Chi-
Chester, England (1983).
77. Spiegler, K.S. and Laird, A.D.K. (Eds.), Principles of Desalination, 2nd
edition, Academic Press, New York (1980).
78. Juda, W. and McRae, W.A., U.S. Patent 2,636,851 (April 28, 1953).
79. Ogami, K. and Kanda, C., Jap. Pat. 1897 (1958).
80. Atsugi, Ichikawa,,T.M. and Yamada, M., Jap. Pat. 2293 (1958).
81. Onogram, K. and Mizutani, Y., Jap. Pat. 91 (1959).
82. Evers, W.L., U.S. Pat. 2,518,420 (1948).
83. Bodamer, G.W., U.S. Pat. 2,597,438 (1952).
84. deKb’rb’sy, F. and Schorr, Y., U.S. Pat. 891,562 (1963).
85. Zschocke, P. and Quellmalz, D., J. Membr. Sci., 22 (1985) 325-332.
86. Bodamer, G.W., U.S. Pat. 2,681,320 (1954).
87. Grot, W.G., Chem. lng.-Tech., 47 (1975) 617.
88. Grot, W.G., U.S. Pat. 3,784,399 (1974).
89. Sata, T., Motani, K. and Ohaski, Y., Perfluorinated exchange membrane
“Neosepta-F” and its properties, in: Ion Exchange Membranes, D.S.
Flett (Ed.), E. Horwood Ltd., Chichester (1983).
90. Seko, M., Yomiyama, A. and Ogawa, S., Chlor-alkali electrolysis using
perfluorocarboxylic acid membranes, in: Ion-Exchange Membranes,
D.S. Flett (Ed.), E. Horwood Ltd., Chichester (1983).
91. Riley, R.L., Lonsdale, H.K., Lyons, C.R. and Merten, U., J. Appl. Polym.
Sci., 11 (1967) 2143-2158.
92. Riley, R.L., Fox, R.L., Lyons, C.R., Milstead, C.E., Seroy, M.W. and
Tagami, M., Desalination, 19 (1976) 113-126.
93. Cadotte, J.E., King, R.S., Majerle, R.J. and Petersen, R. J., J. Macromol.
Sci. Chem., A15 (1981) 727-755.
94. Yasuda, H. and Marsh, H.C., J. Appl. Polym. Sci., 20 (1976) 543.
95. Cadotte, J.E., U.S. Pat. 4,039,440 (1977).
96. Cadotte, J.E., Evolution of composite reverse osmosis membranes, in:
Material Science of Synthetic Membranes, D.R. Lloyd (Ed.), pp.
273-294, ACS Symposium Series 269, American Chemical Society,
Washington DC (1985).
97. Henis, J.M.S. and Tripodi, M.K.,Sep. Sci. Tech., 15 (1980) 1059-1068.
98. Yasuda, H. and Lamaze, C.E., J. Appl. Polym. Sci., 17 (1973).
99. Rozelle, L.T., Cadotte, J.E., Cobian, K.E. and Kopp, C.V., Nonpoly-
saccharide membranes for reverse osmosis: NS-100 Membranes, in:
Reverse Osmosis and Synthetic Membranes, S. Sourirajan (Ed.),
pp. 249-261, National Research Council Canada, Ottawa, Canada
(1977).
100. Cadotte, J.E., King, R.S., Majerle, R.J. and Petersen, R.J., J. Macromol.
Sci. Chem., Al5 (1981) 727-755.
101. Madsen, R.F., Hyperfiltration and Ultrafiltration in Plate-and-Frame Sys-
tems, Elsevier Publishing Company, Amsterdam (1977).
102. Bader, H., Dorn, K., Hupfer, B., Ringsdorf, H., in Advances in Polymer
Science: Polymer Membranes, M. Gordon (Ed.), Springer Vet-lag,
Berlin (1985).
103. Shimidzu, T., Yoshikawa, M.,J. Membr. Sci., 13 (1983) 1-3.
2
Microfiltration
Mark C. Porter
INTRODUCTION
The beginnings of microfiltration (MF) can be traced back into the 19th
century with the synthesis of nitrocellulose in 1845 by Schoenbein. Fick then
used ether-alcohol solutions of the same (collodion) to form the first nitrocel-
lulose membranes in 1855. Even today, the most common polymers used in MF
membranes are mixed esters of cellulose-including cellulose nitrate.
At the turn of the century (1906), Bechhold produced graded pore sizes in
collodion membranes and measured the pore size with the “bubble-point”
method (to be described later).
In the first quarter of the twentieth century, researchers like Bigelow,
Gemberling, Schoep, Brown, Zsigmondy, and Bachmann made significant ad-
vances in the methodology of casting and regulating pore size. It is astonishing
that the art of controlling pore size and microstructure was developed to such a
high degree of sophistication before understanding the mechanism of membrane
formation.
With the help of Professor Dr. Zsigmondy, who was Director of the Insti-
tute of Colloid Chemistry of the University of Goettingen, Sartorius-Werke
Aktiengesellschaft developed a commercial process for making cellulose nitrate
membranes. The membrane was known as the “Zsigmondy Membranfilter.”
Commercial production began in 1927, but sales were largely confined to the
laboratory market.
The first important application of these membranes emerged during World
War II. Gertrude Mueller and others of the Hygiene Institute of the University of
Hamburg developed the membrane for filtering and culturing bacteria. German
water supplies were often devastated or contaminated by bombing raids. A more
efficient method for detecting coliform or pathogenic bacteria was needed. Con-
ventional culturing techniques in liquid or gel-like nutrient media could take up
61
to 96 hours. Mueller discovered that the entire bacterial flora in one liter of wa-
ter could be deposited on a 47 mm diameter membrane within approximately 15
minutes. The membrane could then be placed on top of a nutrient pad allowing
the nutrients to diffuse up through the pores of the membrane to the bacteria
on the surface of the membrane filter. Each bacterium collected would grow
into a colony of thousands overnight. In less that 24 hours of incubation (37”(Z),
the colony would be readily visible and countable by the naked eye.
After the war (19471, the U.S. Joint Intelligence Objectives Agency sent Dr.
Alexander Goetz to Germany to obtain information on membrane filter produc-
tion methods. Goetz visited Membranfiltergesellschaft (Sartorius), and based on
his findings, was awarded a contract by the U.S. Chemical Corps. to further de-
velop the membrane. By 1950, Goetz had improved production methods to
make membranes with higher flow rates and more uniform pore sizes. He also
imprinted grid-lines on his filters to facilitate counting of bacterial colonies.
Based on Goetz’s developments, the Lovell Chemical Company in Water-
town, Massachusetts was awarded further contracts in 1952 to commercialize
production. In 1954, the Lovell Chemical Company sold its membrane manu-
facturing facility to the newly organized Millipore Corporation. Other companies
in both the United States and England then began to exploit the German tech-
nology base to manufacture membrane filters.
Eventually, membranes made of materials other than cellulose nitrate began
to appear:
1962 Gelman Instrument Co. Cellulose tri-acetate

1963 Sartorius Co. Regenerated cellulose
1963 Millipore, Gelman, Sartorius, S&S Polyvinyl chloride and polyamide
1963 General Electric Polycarbonate
1964 Selas Flotronics Silver membrane
1970 Celanese Co. Polypropylene
1970 Gore Corp. Polytetrafluoroethylene
1975 MembranelEnka Polypropylene
1979 Gelman Polysulfone
1980 Millipore Polyvinylidene fluoride
1981 Nuclepore Polyester
1984 Norton Co., Ceraver Alumina
MEMBRANE STRUCTURE AND FABRICATION
All current MF membranes may be classified as either “tortuous-pore” or

“capillary-pore” membranes (see Figure 2.1). The “capillary-pore” structure is
distinguished by its straight-through cylindrical capillaries, whereas the “tortu-
ous-pore” structure resembles a sponge with a network of interconnecting tor-
tuous pores.
The “tortuous-pore” membranes are the most common and include typical
cellulosic membranes and virtually all other polymers. The “capillary-pore”
membranes are currently manufactured commercially only by Nuclepore Corp.
and Poretics Corp. They are available as polycarbonate or polyester membranes.
A unique feature of the “capillary-pore” membranes is that the pore size
Microfiltration 63
can be measured directly with a scanning electron microscope. This is not pos-
sible with a "tortuous-pore" membrane since the pore openings do not corre-
spond to the limiting pore size within the depth of the membrane. This is dem-
onstrated in Figure 2.1 where both membranes (of the same pore size rating) are
viewed at the same magnification. The "tortuous pore" membrane appears to
have some very large pore openings on the surface.
Figure 2.1: Capillary-pore and tortuous-pore membranes.

A look at the open area of the two membranes in Figure 2.1 indicates that
the “tortuous-pore” membranes are more porous-having a porosity over 75%.
The “capillary-pore“ membranes generally have porosities less than 5%. How-
ever, the fact that the latter are ‘/is the thickness of the “tortuous pore” mem-
branes means that the flow rates are often comparable.
Tortuous-Pore Membranes
Phase-Inversion Process. Most tortuous-pore membranes are made by a cast-
ing process known as “phase inversion.” Figure 2.2 is a simplified schematic of a
casting machine which makes cellulose ester membranes. Typically, a casting so-
lution made up of the polymer and a multicomponent solvent system is metered
onto a stainless steel belt or web. The belt passesthrough a series of environmen-
tal chambers usually containing water vapor at elevated temperatures. The more
volatile solvents evaporate and the water vapor precipitates the polymer around
the less volatile solvent which becomes the “pore-former.” Subsequently, (not
shown in Figure 2.2), after the membrane is formed, the residual solvents are
washed out of the pores, surfactants are added, and the membrane is dried.
CELLULOSIC MATERIAL
WlrH SOLVENIS
Figure 2.2: Schematic of casting machine for MF membranes.
There are also a few cases where the “phase-inversion” process is accom-
plished by passing the belt through a liquid water bath to precipitate the poly-
mer as is done with polyvinylidene fluoride.’
Stretching Process. Polytetrafluoroethylene (PTFE) or other chemically-
resistant polymers cannot always be dissolved in an organic solvent. Robert
Gore developed a stretching process for making porous PTFE membranes.2
Gore’s process uses a paste-forming extrusion technique where the extrudate is
stretched at a rate exceeding 2000% per second at an elevated temperature. The
final length of the stretched product is over 50 times the original length. Figure
2.3 shows the nodes and interconnected fibrils of the porous membrane pro-
duced-GoretexTM .
Microfiltration 65
Figure 2.3: Photomicrograph of PTFE membrane (GoreTexTM).
A similar approach has been used by Celanese Corp. to make CelegardTM,

an expanded polypropylene membrane (see Figure 2.4). The stretching creates
elongated pores measuring 0.02 by 0.2011 or 0.04 by 0.4011. The low flow rates
of this membrane limit applications primarily to battery separators and airvents.
Figure 2.4: Photomicrograph of polypropylene membrane (CelgardTM),
Thermal-Phase-Inversion Process. Still another approach to forming porous

membranes out of Polymers not soluble at roOm temperature is to elevate the
temperature until they dissolve in a selected solvent. The resulting solution is
then Cooled in a controlled way until the polymer precipitates around the sol-
vent which serves as the "pore-former" at room temperature. The process is
called "thermal-phase-inversion."
Tracor Hydronautics and Enka (Membrana) have used this process to make
polypropylene MF membranes. It has been speculated that the solvents used
were kerosene or various amines. Enka produces both tubes (5.5 mm I.D./8.6
mm O.D.) and capillaries (1.8 mm I.D./2.6 mm O.D.) in pore sizes of 0.2 and 0.4 Jl
(see Figure 2.5).
Figure 2.5: Photomicrograph of polypropylene capillary membrane (made by

EN KA '5 thermal-phase-inversion process) .
Capillary-Pore Membranes
Many techniques have been proposed for making capillary-pore membranes
including laser beams, electroforming, photochemical etching, and ionotropy to
orient anisotropic gel particles to form ionotropic-gel membranes.3
Glass capillary arrays are now commercially available for laboratory use.
They are formed by assembling a large number of parallel glass capillary tubes,
heating to fuse and draw down to individual capillary diameters of 0.5 ,11.The
bundle is then sliced to form thin discs with a regular capillary array.
Further, a new alumina capillary-pore membrane has just been introduced
by Anotec Separations Ltd. The capillary-pore structure appears to result from
controlled growth of a1umina crystals.
To date, none of these methods has produced submicron capillary-pore
membranes at a reasonable cost in the large areas suitable for industrial appl ica-
.II "
tlons except the track-etch membrane produced by Nuclepore Corp.
The l'track-etch" process was first patented by Price and Walker .4 It consists
of a two-step process (see Figure 2.6) involving a nuclear reactor and an etch
bath. In the first step, a fairly thin film (less than 20,11) of polycarbonate or
polyester is passed between two fission plates of U23S. Thermal neutrons in the
reactor result in massive charged fission fragments bombarding the film. These
particles leave "damage tracks" in the film where polymer chains have been rup-
tured and ionized (see Figure 2.7 ). Theoretically, any dielectric material subjected
to this treatment will be left with "damage tracks'l which will be more vulner-
able to chemical attack than the bulk of the material.
Microfiltration 67
Charged
particles
Pores
A
&!if lxz
Figure 2.6: Track-etch process for capillary-pore membranes.
Figure 2.7: Damage-track in an organic P0lYrt-W.
In the case of polycarbonate or polyester films, the “damage tracks” are sus-
ceptible to attack by caustic (NaOH) solutions in the second step. The caustic
will immediately “strike-through” the track and then begin to etch radially out
from the track. The longer the residence time in the etch tank, the higher the
temperature and caustic concentration, the larger the pore size. Once “strike-
through” has been accomplished, the etch rate is the same in the pores as on the
surface of the film. Thus, when etching a 10~ pore, approximately 10 /A of sur-
face will be removed from the film. Since films thicker than 15 to 20~ will not
allow penetration of the fission particles, there is an upper limit to pore size of
about 12 cc. For example, in Table 2.1, the starting thickness is 18 /.Lwhen etch-
ing a 12 /.I pore. The final membrane thickness, after removal of 12 /.Aof surface,
is only 6 P.
Table 2.1: Capillary-Pore Membrane Specifications
::
Typical Flow Rates f:
Nominal Nominal
Pure Size Pore Denlrify Thickness Water Nitroge3 z
c
(urn) (pores/m 1 (urn) (ml/min/cm2) (l/min/ao1 2
12.0 1 x 105 ::: 6 1400 110 5

10.0 1 x 105 6 1200 75 3
8.0 1 x 106 1':: :x 1100 55 u
2
5.0 4 x 105 1000 50 2
3.0 2 x 106 1000
2.0 2 x 106 E ix 450 5:
1.0
0.8 2
3 x 107 1.0 ix 300 zx
0.6 3 x 107 1.0 10 200 15
0.4 1 x 106 1.0 10 1'5" 10
::: 3 x 106
108 1.0
0:s 10
5 8.0 :::
0.08 6 x 108 X:i: : 2.0 1.0
0.05 6 x 108 0.2 0.5
0.03 Experimental material--firm 8pecificatiom not yet established
0.015 Experimental material--firm specification8 not yet crtablished
Microfiltration 69
The density of pores (number of pores/cm2) is determined by the residence

time and/or power in the nuclear reactor. Higher pore densities result in higher
flow rates. However, for each pore size, there is an upper limit of pore density
above which the membrane becomes intolerably weak. In general, porosities
above 10% result in low strength membranes. In an attempt to maximize flow
rate and still maintain good strength, the pore density is increased for the smaller
pore sizes in Table 2.1. However, pore densities above 6 x 108 pores/cm2 require
longer residence times in the reactor which tend to scorch the film with the
longer exposure to radiation. Therefore, at pore sizes of 0.1 JJ.and below, the
membrane thickness is reduced by 50% to 5 JJ.to increase the flow rate by a fac-
tor of 2.
Since the irradiation step is a random bombardment process, there is a
small probability that two tracks will be adjacent to each other forming doublet
pores, triplets, etc. (see Figure 2.8). Obviously, two or three pores which run
through the entire thickness of the membrane, adjacent and parallel to each other ,
can have a marked effect on the retention characteristics of microorganisms. The
probability of this occurrence is further reduced by collimating the fission frag-
ments so that they bombard the membrane at angles between 0° to 29° with an
average angle of attack of 10° (see Figure 2.9) .Th is is of I it tie hel p when the
pore size is large as in Figure 2.10. Here, two 5 JJ.pores diverge, but not enough
to form two distinct pores over a th ickness of only 10 JJ..However, at a 0.2 JJ.
pore size, which retains the smallest bacteria, the L/D ratio for the pore is much
higher 10/0.2 = 50.
Even so, when one considers the number of pores in one cm2 (3 x 108), the
statistical probability of pore overlap on both faces of the membrane is very
high.s,6 Further, in a 10-inch pleated cartridge, there are 20 square feet of mem-
brane containing 6 x 1012 pores. There is an overwhelming probability that many
doublets, triplets, quadruplets, etc. are present. Apart from damage to the mem-
brane during manufacturing, this is the underlying reason why Nuclepore will
never be able to make a "bubble-pointable'l cartridge out of this membrane.
Figure 2.8: Doublet and triplet pores in capillary-pore membrane.

P~LYCARB~NATE
rl--
! I
FllM i
OPEN AREA: -15z

PORES/cm2: 20-30 MILLION 00
- 100I-
Figure 2.9: Angle of pores in capillary-pore membrane.
Figure 2.10: Doublet pore in 5.0 micron capillary-pore membrane.
PORE SIZE DETERMINATION
Challenge Tests
It is easy to measure the pore size of capillary-pore membranes with a scan-
ning electron microscope, but tortuous pore membranes are more difficult.
Since the first applications of MF membranes were in the filtration and cul-
turing of microorganisms, membranes were characterized by their ability to re-
tain specific organisms 100%. As shown in Figure 2.11, Pseudomonas diminuta
was thought to have a minimum diameter of 0.22 cc, and this became the pore
size rating for those membranes which retained Pseudomonas completely. Like-
wise, Serratia marcescens was thought to have a diameter of 0.45 j,r, etc. In
fact, a significant variation in size has been documented: up to 30% in diameter
and 67% in length.
Microfiltration 71
.
Red Blood Cell \. ’ ‘.
Yeast
1 5.0 pm
Coliform
Serra tia
marcescens
Pseudomonas-
Figure 2.11: Challenge organisms for various pore sizes.
Every manufacturer still challenges tests measuring the “Beta ratio” (the
number of organisms challenging the membrane f number of organisms passing
through). Normally, a membrane is considered good if challenged with enough
bacteria to plug the pores (typically lOs-10s organisms per cm2 of filter area)
and none pass. However, some manufacturers are more conservative than others-
starving bacteria before the test to challenge the membrane with the smallest
viabile organism possible.
Other particles may also be used to test the membrane. Monodisperse latex
spheres (produced by Dow Chemical) in concentrations of 106-10’ particles per
ml are sometimes used. Normally, a second membrane is used downstream of the
test membrane to collect particles which pass. The number of spheres collected
are counted with the use of a scanning electron microscope. This is much more
tedious and less accurate than the bacteria challenge test where bacteria col-
lected on the second membrane may be grown into colonies visible to the na-
ked eye. Alternatively, automatic particle counters are used before and after the
membrane, but the method is not as sensitive and is limited to particles over 0.5 /J
in size.
Bubble-Point Test
An easier test which is also nondestructive is the “bubble-point” test. The
maximum pore size may be determined by measuring the gas pressure required
to overcome the capillary forces holding a wetting liquid within the pores (see
Figure 2.12).
ZERO INCREASING BUBBLE POINT

PRESSURE PRESSURE PRESSURE
Figure 2.12: Procedure used in determining bubble-point.
The bubble point equation is essentially the capillary rise equation:
4y cos 0
(1) P= -yj-
where P = bubble point pressure

d = pore diameter
7 = surface tension of the liquid filling the pores
8 = liquid-solid contact angle (6 = 0” for perfectly wetting fluids;
0 > 90” for nonwetting fluids).
For water, with a surface tension of 72 dynes/cm, and a contact angle of O”,
equation (I) reduces to:
where P is in psi and d is in microns.
For isopropanol, which wets almost all polymers, the equation becomes
(with the same units) :
(3) p= a
d
Microfiltration 73
First, the membrane is completely wetted with liquid and then a gas pres-
sure is applied to one side. As the gas pressure is gradually raised (see Figure
2.12), there will be no gas flow through the pores until capillary forces are over-
come releasing liquid from the pore. Obviously, from equation (I), the first gas
bubble will emerge from the largest pore-where the capillary forces are lowest.
The pressure at which this occurs is called the “bubble-point” pressure. The
maximum pore size may then be calculated.’
If the gas pressure is raised further, more pores will begin to bubble until the
liquid is displaced from the smallest pore and the gas-flow through the mem-
brane equals that through a dry membrane. A pore size distribution may be cal-
culated from these data: Indeed, an ASTM test-method’ outlines the procedure:
(1) Measure the dry air flow and plot versus pressure (see Figure 2.13).
(2) Wet the membrane and raise the air pressure until the first air-
bubble appears; this is the “bubble point” and may be used to cal-
culate the maximum pore size.
(3) Continue to raise the air pressure and measure airflow through the
wet membrane as a function of pressure.
(4) Continue to raise the pressure until the airflow through the wet
membrane equals that through the dry membrane at the same pres-
sure. At this point, all liquid has been expelled from the membrane
and the pressure may be used to calculate the minimum pore size.
(5) To calculate the mean-pore size, ASTM method R316 constructs a

line equal to one-half of the dry air flow through the membrane.
The pressure at which the wet airflow intersects this line is the
point at which the liquid has been expelled from one-half of the
pores and may be used to calculate the mean pore size.
PRESSURE-
c--PORE DIAMETER
Figure 2.13: Procedure used in determining maximum, mean, and minimum

pore size.
There is one complication in the bubble-point test referred to as “diffusional-

flow.” A small amount of gas-flow can result even through a pore is filled with
liquid. The gas dissolves in the liquid in the pores at high pressure, diffuses
across the liquid-filled pore in solution, and comes out of solution on the low-
pressure side of the membrane. In practice, “diffusional-flow” is not even de-
tected when small membrane areas are involved. Even for large areas, it is easily
distinguished from the much larger gas-flow at the bubble point.
It is helpful to estimate the amount of flow expected due to diffusion:
J = m2 (Dl-l) &
(4)
4 L
where J = gas flow rate per unit area of membrane

N = number of pores per unit area
d = pore diameter
D = diffusivity of the gas in the liquid
(D = 1.64 x 1O-’ cm2/sec for N2 in water at 20°C)
H = the solubility of the gas in the liquid
(H = 6.9 x IO-‘gm mols/atm/cm3 for N2 in water at 2O’C)
AP = the pressure differential across the membrane
Q = the tortuous path length through the pore and across the
membrane
Figure 2.14 is a convenient chart to use in estimating “diffusional-flow.”

The chart was originally constructed for use with capillary pore membranes with
a thickness of 10 cc. However, since the parameter Nrd2/Q in equation (4) is ap-
proximately the same for capillary and tortuous pore membranes, the chart may
be used for both. Table 2.1 should be used to identify the density corresponding
to the pore size in question. For example, a 0.2 (u pore size has a density of
3 x lo* pores/cm2. Therefore, both capillary and tortuous pore membranes would
be expected to show a diffusional-flow of about 0.1 ml/min/ft2/psi. For a 10
square foot pleated cartridge tested at 30 psi, the diffusional flow rate should
only be 30 ml/min. (This can be readily measured with a liquid-filled inverted
graduate or burette-see Figure 2.15).
The estimates obtained from Figure 2.14 will be slightly low for tortuous
pore membranes at pressures near the bubble point. This is because the pore
opening is generally larger than the pore neck (see Figure 2.16). At higher pres-
sures, the liquid will flow from the pore entry region until it reaches the smaller
neck where it will stop unless the pressure exceeds the bubble point. Thus, the
diffusion path through liquid in the pore is reduced, resulting in higher diffu-
sional-flow rates by equation (4). In practice, this increase is mitigated by the
fact that the membrane support also offers additional resistance to diffusion.
As previously noted, Nnd2/Q is the key parameter in equation (4).
Membranes of equal porosity and thickness will show the same “diffusional-
flow.” This is why the so-called “forward-flow” test advocated by some manu-
facturers is so misleading. They claim they can correlate bacteria retention with
the diffusional-flow measured. Yet, most pore sizes of tortuous-pore membranes
have approximately the same porosity and thickness. Equation (4) makes it clear
Microfiltration 75
that a “forward-flow” test cannot distinguish between a 1.2 /J membrane which

passesPseudomonas diminuta quantitatively, and a 0.22 /.Lmembrane which will
retain it quantitatively-if the membranes have the same porosity.
Valid for pressures

below bubble point
Assuming membrane
thickness of IO )rrn
Pore size, pm
Figure 2.14: Diffusional-flow for various pore sizes and pore densities.
Figure 2.15: Method for measuring diffusional-flow rate.
Figure 2.16: Bubble-point in membranes with pore openings larger than the
internal pore size.
Microfiltration 77
Mercury Intrusion Test

Mercury is a non-wetting fluid for most materials. Because the contact angle
(0) is 180°, cos = -1, and pressure is required to force mercury into the pores-see
equation (1). We speak of mercury “intrusion pressures”; these are quite high
due to the high surface tension of mercury (476 dynes/cm). Thus, for a given
pore size, the pressure required to force mercury into the pores is almost seven
times greater than the pressure required to expel water from the pores.
The technique places as much membrane area as possible into a chamber
which is evacuated. Mercury is then admitted into the chamber as the pressure
is slowly elevated. The incremental volume of mercury added for each increase
in pressure is measured precisely.
The greatest ambiguity is that the volume of mercury penetrating dead-
ended “pits” is registered along with that penetrating “pores.” Further, the high
pressures required may collapse some pores. Nevertheless, mercury porosimetry
can be helpful in confirming the pore-size distribution obtained by other meth-
ods such as bubble-point.
Other Methods
A number of other techniques have been used to determine pore size, but
they are not accurate enough to be used as standard characterization tools.
Flow Permeability Test. A rough estimate of an equivalent mean pore size
may be made by measuring the porosity (E) and permeability (J) of the mem-
brane. Assuming laminar flow, the Hagen Poiseuille relationship may be modi-
fied for a porous membrane:
4
(5) J=_
128~ L
where J = volumetric liquid flow rate per unit area of membrane

N= number of pores per unit area
d = pore diameter
AP = the pressure differential across the membrane
cc = viscosity of the liquid
P = the length of the pore
The porosity (e) of the membrane is the same as the void volume of the
membrane and may be determined by measuring the difference between the
wet and dry weights of the membrane. Since
E d 2AP
J= -
32~1
or
(8) d-
All the variables on the right side of equation (8) are experimentally measurable
except for II, the length of the pore. Obviously, II is related to the thickness of
the membrane by a “tortuosity factor.” The problem is that we cannot meas-
ure the “tortuosity factor” with any degree of certainty.
Even the case of “capillary-pore” membranes where R is only slightly larger
than the thickness of the membrane, inertial losses, especially “front-and-back-
face” losses, may be significant.
Smoke DOP Test. The classic method’ for determining the filtration effi-
ciency of various air filters is to challenge the filter with a liquid aerosol of di-
octyl phthalate (DOP). Typically, the aerosol particles are 0.3 p in diameter.
Although the method has been used to characterize MF membranes for the
filtration of gases, the retention of a liquid aerosol does not correlate well with
the retention of particles suspended in a liquid.
BET Adsorption Data. A wealth of information about the size and shape of
pores may be obtained from adsorption isotherms where the mols of nitrogen
adsorbed on the membrane are measured as a function of pressure. However, the
use of this techniques is not widespread due to the tedious regimen required in
gas adsorption measurements. Further, the hysteresis effects make conclusions
about pore-structure ambiguous.
RETENTION CHARACTERISTICS
Bacteria Retention and Bubble Point

In 1974-1976, Wallhausser wrote several controversial papers,“r” ,showing
that MF membranes may pass microbes at high challenge levels (see Table 2.2).
Previously, membrane manufacturers had insisted that their membranes were
“absolute” and many were skeptical of Wallhauser’s data.
Shortly thereafter, other data began to appear in the literature confirming
Wallhausser’s findings. Eventually the manufacturers collected their own data
which showed that no membrane is truly “absolute,” for example, Table 2.312
shows that filtration time as well as challenge level is a factor.
Some researchers used the term “grow-through” to explain the results. The
idea was that bacteria trapped in the neck of the pore subdivide during growth
so that the new organisms emerge through the neck to the other side. However,
calculation of the maximum pore-size from the bubble-point shows that there
are some pores larger than challenging organism. For example, the theoretical
water bubble-point of a membrane with a membrane with a maximum pore size
of 0.2 I-( is calculated from equation (2) to be 200 psi. Most 0.20 /.r cellulose es-
ter membranes (water wet) have bubble-points between 45 and 60 psi, which is
only one-fourth of the theoretical value.
Table 2.2: Wallhausser’s Data on Bacterial Break-through”
Pore size 0.2 pm (flow-rate: lOOmI water in 22 see) and 0.45 pm (flow-rate: 100 ml water in 13 see).
I
Initial bacterial
IOVml I IV/ml 1VVml
concentration+) I
Filtrate No. of bacteria after filtntion
0.2pm I 0.45 urn I 0.2 urn I 0.45 rm I 0.2 firn I 0.45 pm
100 ml
200 ml
1000 ml
Filtration timc for 1OOOml 1 6’52” I 2’27” 2hl2 2’30” 3h 15
I I g’
l) Pseudomonas diminuta ATCC 19 146.

Table 2.3: Bacterial Penetration with Time12
Time, Bacteria
Filter No.” Rating hr Recovered
PA 5-2488 0.2 pm 24 O/L

48 O/L
168 62/L
312 750/L
PA 5-248D 0.2 Ccm 72 O/L
119 O/L
170 16/L
265 20/L
314 19/L
CET-1 0.22 pm 24 I9 total
CED-1 0.22 pm 41 I9 total
162 260 total
216 IO00 total
CET-2 0.22 pm 24 I total
CET-3 0.22 pm 24 0
PA 538 0.2 pm 24 1I total
48 1000 total
PA 562 0.2 Itm 89 0
PA 595 0.2 pm 90 0
* PA - nylon 66. CE = Cellulose ester.

Note: All tests run to plugging (40+ psid).
Input levels ranged from I O4to 1O5bacteria/
liter.
The manufacturers early recognized the discrepancy between theoretical

and experimental bubble points. They introduced an experimental constant into
equation (1) called a “shape-factor” which supposedly took into account the
fact that tortuous pores are not cylindrical in shape. However, a shape factor of
0.25 cannot be justified from the capillary rise equations, as any good textbook
on surface chemistry will attest. Indeed, for noncircular pores, equation (1) is
adjusted to:
p = wCos@
(9)
A
where p is the perimeter of the pore and A is its area. Thus, for pores of equal
areas, the one with an irregular shape will yield a higher bubble point than a cir-
cular pore!
Still, other researchers explained the discrepancy between theoretical and
experimental bubble-points with surface tension depression. It is true that most
manufacturers treat their membranes with wetting agents. It is conceivable that
these surfactants can dissolve in the test fluid and lower the surface tension.
Microfiltration 81
However, the discrepancy between theoretical and experimental values exists

for all test fluids. For example, it is impossible to postulate a depression in the
surface tension of isopropyl alcohol from 22 to 5 dynes/cm.
It is apparent that the bubble point is telling us that a 0.2 /..Ipore size mem-
brane has at least one pore up to 0.8 /J in diameter. This is confirmed by mer-
cury porosimetry data (see Figure 2.17).
50
45
40
35
30
2s
IS
10
0.4 0.5 0.6 0.7 0.8 0.9 I.0

Port-Size
inwn
Figure 2.17: Distribution of pore sizes in 0.45 micron membrane (by mercury
porosimetry).
To verify the theoretical equation for bubble point, a series of 0.2 p capil-
lary pore membranes were made with extremely low pore densities. The lowest
densities virtually eliminated all doublets and triplets and yielded bubble points
equal to the theoretical value. Though the low density rendered the membrane
useless for applications requiring a reasonable flow rate, it nevertheless demon-
strated that a membrane can be made where a direct measurement of the pores
(with S.E.M.) agreed with that calculated from the bubble-point.
From the above, it is clear that given enough time and bacteria, some will
find a leak-path through the MF membrane.
There is, however, a long history of successful membrane usage in the steri-
lization of fluids by filtration. Indeed, for all practical intents and purposes, the
membrane has been “absolutely retentive.” There are few applications which
subject the membrane to the challenge levels used by the manufacturers in their
quality control (typically, 108-10’ organisms per cm2 of filter area). Such a
challenge provides more than one organism per square micron of filter area and
often clogs the membrane.
It is true that in ultrapure water loops, where the membrane is in service for
many months, the accumulation of bacteria on the membrane can reach these
levels. This explains why there can be more organisms in the filtrate than can be
detected in the feed stream.
Retention of Deformable Particles

Some researchers have argued that, under pressure, bacteria will passthrough
pores of smaller diameter due to deformation. Most microbiologists are doubtful
that Pseudomonas diminufa will deform to this degree. Nevertheless, it is recog-
nized that larger microorganisms can deform.
For example, Figure 2.1813 demonstrates the transmission of normal de-
formable red cells through various capillary-pore membranes. After hardening
the red cells, they will not pass anything below a 9 /.I pore size.
The author has seen data on the passage of yeast particles through pore sizes
under 3 1-1which has suggested passage by deformation.
I I I I I i i
NORMAL IIC HARDENED I8C
I I I I I I I
1 4 6 8 IO 12 I4
FILTER PORE DIAMETER (p)
Figure 2.18: Transmission of red blood cells through capillary-pore membranesI
Retention by Adsorption
Lukaszewicz et all4 has postulated an “adsorptive” mechanism to explain
the excellent retention of bacteria by membranes with some pores larger than
the microorganism. He contends that the adsorptive sequestrations of the mem-
brane are more important than the geometric restraints of sieving.
Davis et al” investigated the retention of 0.05 and 0.005 ~1 Au colloids by
“capillary-pore” and “tortuous-pore” membranes. Table 2.4 shows clearly that
the “tortuous-pore membranes” retain particles much smaller than the rated
pore size. Indeed, a 5 p pore size will retain 60% of the 0.05 /J colloidal particles
and 18% of the 0.0005 /J colloid. On the other hand, “capillary-pore” membranes
retain less than 1% of either. “Tortuous-pore” membranes have 25 to 50 times
more internal surface area for adsorption than “capillary-pore” membranes, and
the tortuous path also results in a greater likelihood of small particles contacting
the pore-wail.
Microfiltration 83
Table 2.4: Retention of Au-Colloids on Capillary-Pore and Tortuous-Pore

Membranes
Colloid Size (urn) 0.05 0.005

O.lum Nuclepore 0.2
O.lvm Cellulosic 9% 8.2
0.4Um Nuclepore 1.3
0.45um Cellulosic 46.9 120:;
l.Ovm Nuclepore 0.7 0.3
1.2pm Cellulosic 46.5 6.7
3 .Ovm Nuclepore 0.4
5.0ttmCellulosic 59.3 A92
These results suggest that a “tortuous-pore” configuration is best for “clean-

up” applications where the removal of all particles from the process stream is de-
sired. On the other hand, the “capillary-pore” configuration is best for fraction-
ation of particles. For example, capillary-pore membranes have been used in
fractionating silver colloids to improve resoltuion on photographic films.
For dilute process streams, product may be lost via adsorption on the mem-
brane. The “recovery” of this product may be improved by pretreating the mem-
brane such that most of the adsorption sites are occupied. For example, in the
data of Hahn et all6 (Table 2.5), polio virus adsorption on cellulosic “tortuous-
pore” membranes was significantly higher than that on polycarbonate “capillary-
pore” membranes. (i.e., The virus recovery is low due to adsorption.) The recov--
ery was improved from 5 to 76% by pretreating the membrane with a beef ex-
tract solution.
Table 2.5: Recovery of Poliovirus in Retentate or Filtrate after Filtration on

Various Filters
SuspendinS r&am*
Filter She @II)
BSS BSA(1%) BE (O.OOl?&)
BE WI%1 BE (0.3%%)
.2
Silver............. 60 75 98 86
Silver. ......
.45 ; : 78
22
cellUlosc.......... <5 <5 <5 7 61 76
......
Cellulose... .45 <J <5
Polycarbonate. .5 73 80 7; 80
lBSS, Hanks balanced salt

solution; BSA, bovine albumin in distilled water; BE; beef extract in
BSS.
Retention by “Charged” Membranes

In certain aqueous streams, the retention of charged species may be enhanced
without sacrificing the filtration rate by using charge modified membranes. Since
most particles are negatively charged, these membranes are usually positively
charged. Normally a crosslinking epoxy resin with cationic functional groups
(e.g., quarternary ammonium groups) is used to impart the charge. The com-
bination of electrokinetic adsorption with mechanical sieving is claimed to result
in significant improvements in retention.
The most significant claim is for the removal of “pyrogens” (endotoxins).

Table 2.6 ” compares the pyrogen removal efficiency for a positively charged
nylon 0.2 p pore size membrane, ZetaporTM, and a conventional cellulosic 0.22 p
membrane. Normally, a 10,000 molecular weight cut-off UF membrane is re-
quired to remove pyrogens. (These membranes have an equivalent pore-size of
30 A or 0.003 or). As expected, the conventional MF membrane shows no reten-
tion whatsoever, but the positively charged membrane shows better than 97%
retention.
Table 2.6: Pyrogen Removal by “Charged” Membranes
~1
Racedure: E. coli purifiedcndotoxinwasaddedto 0.9% NaCl solutionat pH 6.7

and passedthrough filters containedin a25mm filter holder.
It is also claimed that positively charged 1.2 p pore size MF membranes will
retain Pseudomonas diminuta. Data from Pall Corp.‘s on their 1.2 p Ne6
PosidyneTM membrane show better than 99.99% retention of Pseudomonas.
Their conventional 1.2 p ULTIPOR Nh6TM membrane (without the charge) re-
tained less than 50% of the Pseudomonas. In both cases, challenges between
2 x lOlo and 7 x 1Ol2 bacteria per square foot of filter area were used. This
means that comparable removal of Pseudomonas can be achieved with more than
10 times the flow rate of a 0.2 p pore-size membrane.
Further, if all of the particles retained are considerably smaller than the
rated pore size (say less than 25%) and no larger particles are present, the mem-
brane does not clog even with the processing of very high volumes. The small
particles do not “bridge” and cause only a slight change in the pore diameter.
The difficulty is that, with use, the filtration efficiency of these positively-
charged membranes begins to drop. As the membrane adsorbs more and more
negatively-charged particles the adsorption sites (cationic functional groups) are
used up. Just as a saturated ion-exchange resin is incapable of further retention
of ions, the membrane begins to pass negatively charged particles. Figure 2.19l’
shows the decline in retention for 0.2 p latex particles.
Similarly, high flow rates (pressure drops) through the membrane may strip
the negative particles away from the membrane (see Figure2.20).19The hydraulic
shear forces have overcome the electrostatic attraction.
For critical applications, one cannot afford to guess when the adsorption
capacity is used up. However, in some cases, a positively charged membrane of
larger pore size than the final membrane may be used as a prefilter, decreasing
the load on the “absolute” filter and extending its life. For example, it is re-
ported that a 3 p positively charged prefilter can extend the life of a 0.2 ccfinal
membrane filter by a factor of 2- to 6-fold.‘s
Microfiltration 85
ASYMMETRIC POLYSULFONE
3s%=QT@
CHARGED NYLON
B.
LATEX SIZE 0.196 MICRONS
COUNT 5 x 1O’O PARTICLES/ML
1 I I I I t
2 4 6 6 10 12
VOLUME, ml
Figure 2.19: Retention of 0.2 micron latex particles by “charged” and uncharged
membranes as a function of volume filtered.
l-
0
ASYMMETRIC POLYSULFONE
I -
CHARGED NYLON
LATEX SIZE 0.196 MICRONS

COUNT 6 x 10’0 PARTICLES/ML
I I I
0 10 20 30
PRESSURE
Figure 2.20: Retention of 0.2 micron latex particles by “charged”and uncharged

membranes as a function of filtration pressure.
Alternatively, two positively charged membranes may be used in series. The

adsorption capacity of the upstream filter will be exhausted first. When the ef-
fluent from this filter begins to show passage, it is replaced by the downstream
filter. A new filter is then placed in the downstream position.
Negatively charged membranes are also available. The fractionation wpabil-
ity of “tortuous-pore membrane ” is enhanced considerably by using filters with
a zeta potential of the same sign as the particles (usually negative). It should be
noted that untreated nylon membranes have a negative zeta potential at pH val-
ues above 6.5 (see Figure 2.21).18
-25
-30
\,
PH
Figure 2.2 1: Zeta potential of positively charged and uncharged nylon mem-
branes.
Aerosol Retention
The filtration of particles in a gas stream can be quite different from the fil-
tration of the same particles in a liquid stream. The three mechanisms of aerosol
particle retention may be illustrated from the data of Spurny et aI” in Figures
2.22 and 2.23. The U-shaped curves are characteristic of the efficiency of aero-
sol particle collection as a function of particle size. However, “capillary-pore”
membranes have a deeper minimum in the curves than do “tortuous-pore mem-
branes.”
Microfiltration 87
0.01 0.1 1.0

Porticiesize r,pm
3. R=l.O;m
a R = 2.5 $T s = 21 g/cc
5 R=40c,m t
R = Pore Radius; r = Particle Radius; q = Face Velocity of Gas;
5 = Particle Density; E = Particle Collection Efficiency.
Figure 2.22: Aerosol retention on a capillary-pore membrane as a function of

particlesize and pore size.
Particle size r, pm
1. q = 0.1 cm/se.
R = 4.0 *In
2. q = 1 .o cm/set
3 4 = 5 0 cm/s-x. s = 21 g/cc
4. q = 25 0 cmlsec t
R = pore Radius; r = Particle Radius; q = Face Velocity of GaSi
s = particle Density; E = Particle Collection Efficiency.
Figure 2.23: Aerosol retention on a capillary-pore membrane as a function of

particlesize and gas velocity.
Direct Interception. This mechanism is common to both liquid and gas fil-
tration. The particle is simply too large to pass through a pore. Both Figure 2.22
and 2.23 show that 100% of all particles larger than the pore radii (R) are cap-
tured.
Inertial Impaction. Particles smaller than the rated pore size can still be cap-
tured by "inertial impaction" if they have enough mass to continue in a straight
line when the flow streamlines bend to go through the pores. The result is that
the particles impinge on the rim of the pore or on the pore-wall near the pore-
entrance (see Figure 2.24). The smaller pore sizes increase capture by inertial im-
paction (see Figure 2.22) because there is less likelihood the particles will be
able to follow the flow lines all the way through a narrow pore. Also, higher gas
velocities impart additional momentum or inertia to the particles, thereby in-
creasing the number of particles captured (see Figure 2.23).
Figure 2.24: Latex particles captured on a capillary-pore membrane by inertial

impaction.
Diffusional Deposition. Very small particles will not be captured by inter-

ception or by inertial impaction since their mass is too small. However, these
Microfiltration 89
smaller particles have greater Brownian motion and greater diffusivities. In gases,
the diffusivity is large enough to result in deposition on the pore-wall (see Fig-
ure 2.251. In this case, smaller pore sizes increase the probability of particle im-
pact with the pore wall because of the shorter path length (see Figure 2.221.
Lower gas velocities favor particle deposition because of a longer residence time
in the pore (see Figure 2.231. For dry gases, electrostatic forces can also boost
the capture of these small particles.
Figure 2.25: Silver (Ag) particles captured by a capillary-pore membrane by

diffusional deposition.
“U-shaped retention curves” are seldom seen in liquid filtration. The parti-
cle diffusivities in the higher viscosity fluid are much smaller. As a result, the re-
tention efficiency does not increase with smaller particle sizes.
For “tortuous-pore membranes” the minimum in the curves of Figures 2.22
and 2.23 is much less pronounced. This is because the tortuous path results in
more and smaller particles captured by inertial impaction. Further, the longer
path length through the pore results in more and larger particles captured,by
diffusional deposition.
Thus, if the objective is to capture as many particles as possible on the mem-
brane, “tortuous-pore” membranes are preferred.
On the other hand, for some specialized analytical applications where frac-
tionation of the aerosol particles is the objective, “capillary-pore” membranes
are preferred. For example, it has been found that an 8 /.r “capillary-pore” mem-
brane will collet air-pollution particles that are normally deposited in the upper
respiratory tract (nasopharynx.)” Air sampling stations have used this membrane
in the first stage; particles passing are collected on a tortuous-pore membrane in
a second stage to simulate what is deposited in the lungs.
MEMBRANE PLUGGING AND THROUGHPUT
Once membranes with the appropriate retention characteristics have been

identified, their cost and “throughput” (volume processed before plugging) will
often dictate the membrane of choice. Further, those process schemes and oper-
ating variables which maximize filter life and throughput need to be considered
to improve the economics of filtration.
Membrane Dirt-Loading Capacity

It should be obvious from the section on “Retention by Adsorption” that
“tortuous-pore membranes”, with 25 to 50 times more internal surface area for
adsorption than “capillary-pore” membranes should have higher “throughputs”.
That this is the case is shown in the throughput curves of Figure 2.26. The data
were run at constant flow rate and show the increase in pressure drop across 47
mm filters (AP) with increasing volumes of fluid passing the filter. “Tortuous-
pore” cellulose ester (CE) and polytetrafluoroethylene (PTFE) membranes
showed the highest throughputs. “Capillary-pore” polycarbonate (PC) mem-
branes plugged most rapidly. One must use more area of the “capillary-pore”
membranes to equal the same dirt-loading capacity of a “tortuous-pore” mem-
brane.
Prefilters
The use of positively-charged open-pore membranes as prefilters for a final
“absolute” membrane has already been mentioned. In general, a prefilter serves
to take the load off the final filter. The best prefilters will have a high internal
surface area (high dirt-loading capacity).
It is well known that fibrous depth filters provide enormous dirt loading ca-
pacity compared with membrane filters. Using this kind of filter media as a pre-
filter for the final membrane filter often provides an ideal combination (see Fig-
Microfiltration 91
ure 2.27). Most of the contaminants are removed by the prefilter but the final
membrane serves as the ultimate barrier trapping all particles leaking through the
prefilter-including fibers which “sluff-off” the fibrous media due to “media-mi-
gration”.
7ffROUGUPU7 -ML
Figure 2.26: Tap-water through-put for polycarbonate (PC) capillary-pore

membranes and cellulose ester (CE), polytetrafluoroethylene (PTFE) totruous-
pore membranes.
Depth
Pref ilter ~
Membrane Filter
Figure 2.27: Depth prefilter over final membrane filter.
The smaller the fiber diameter used in the prefilter, the greater the surface
area for adsorption of particles and the better the retention of small particles.
In the sixties, asbestos fibers were recognized as the best prefilter media. The
individual fibrils were smaller than 0.01 p and they had a positive zeta potential.
However, when it was suspected that asbestos fibers presented a health hazard,
fine diameter glass and synthetic polymer fibers were substituted. Unfortu-
nately, neither media equals the performance of asbestos. Glass fibers are avail-
able in the finest diameters, but some users are fearful they may represent a sim-
ilar health hazard. The trend has been to use polypropylene or polyester fiber
prefilters. Melt blown or spun-bonded fibers are available in diameters near 1 p.
Multilayers of these media with appropriate calendering have resulted in surpris-
ingly efficient prefilters.
The selection of the optimum prefilter/membrane combination is best done
experimentally on the user’s process stream. The manufacturers give some guid-
ance as to which prefilters should be used for a given membrane pore-size, but
unfortunately, there is no standard rating system for prefilters.
The experimental procedure used by the author places various prefilters on
top of the selected membrane in 47 mm holders which are run in parallel on the
process stream to be filtered. Usually, the test is carried out at constant flow rate
and the rise in pressure drop is recorded versus the volume processed until the
membrane plugs. The combinations providing the highest throughput are then
evaluated with respect to cost of the media.
The search for the optimum prefilter media is often facilitated by testing
the membrane and prefilter separately after the combination has reached the
limiting pressure drop. If the final membrane is plugged but the prefilter shows a
low pressure drop (see Figure 2.281, the prefilter is too coarse and a more reten-
tive prefilter should be selected to protect the final filter.
Limiting
VOLUME Batch Volume

Figure 2.28: Inadequate prefilter (too coarse) for final membrane filter.
On the other hand, if the prefilter shows a high pressure drop and the mem-
brane a low pressure drop (Figure 2.291, the prefilter is doing a great job protect-
Microfiltration 93
ing the final filter, but additional dirt-loading capacity is required. This may be
accomplished by adding additional prefilter area. In some cases, a more open
prefilter can be used before the tighter prefilter. If the final membrane has a
large pore-size, it is also possible that the prefilter may be too tight.
Limiting*P --_-----__
AP
VOLUME Batch Volume
Figure 2.29: Inadequate prefilter area for final membrane filter.
The optimum match between prefilter and membrane will usually show
both prefilter and membrane plugged so that neither one has carried the com-
plete load by itself (see Figure 2.30).
Limiting AP -- - -- -----_
AP
I
VOLUME Batch Volume
Figure 2.30: Optimized prefilter/membrane combination.
A series of graded prefilters can provide the highest throughput. In Figure

2.31, tests were run on serum at constant pressure showing the volume proc-
essed at various times. The declining slope of the curves reflects the declining
flow rate with time. More open membranes were used as prefilters for the final
0.45 /.r pore-size membrane. PlOO is a glass-fiber prefilter normally matched with
a 1 /.r pore-size final membrane. Obviously, the increases in throughput achieved
by additional media must be weighed against media costs.
)-
I-
I I I I
lo 46 60 &I Jr
TIME (MINUTES)
Figure 2.31: Serum through-put as a function of various combinations of pre-
filters.
Some MF membranes have a variation in pore size from upper to lower face
(see Figure 2.32). We call this “anisotropy”. At least one manufacturer inten-
tionally makes an anistropic MF membrane. The idea is to provide a built-in
prefilter with the more open pore-size upstream of the finer pore-size. Indeed,
an improvement in throughput is achieved, but usually at the expense of reten-
tion. When using discs of anisotropic membranes, orientation can have a dra-
matic effect on throughput. One manufacturer puts a note in every box of mem-
brane: “Use this side upstream”.
Microfiltration 95
Cross-section of the BTS Polysulfone membrane.

XSOO magnification
Figure 2.32: Crossosection photomicrograph-anisotropic tortuous-pore memo

brane.
Effect of Filtration Rate on Throughput

Irregardless of the type of filtration media, it is universally true that lower
filtration rates result in higher throughputs. It should be obvious that two iden-
tical filters run at two different flow rates will show different pressure drops
across the filter. At half the filtration rate, the pressure drop will be one-half as
well. For the same volume processed, and the same degree of pore-plugging, the
filter run at 50% of the flow rate of the other will show 50% of the pressure-
drop as well. If the limiting pressure drop is 30 psi (as in Figure 2.33), the fil-
ter run at 50% of the flow rate will process a larger volume before its pressure
drop reaches the limit.
For any type of pressure filtration, the filtration rate per unit area or flux
(J) will be proportional to the pressure drop (D.P) across the membrane or fil-
ter (i.e., the driving force) divided by the resistance to flow. The resistance term
consists of two parts: the resistance of the cake which accumulates on the up-
stream surface of the membrane ( Rc ) and the resistance contributed by the
membrane itself (Rm ).
In conventional filtration of particulates:
(10) J z ~p
Rm + Rc
The resistance of the membrane, ( Am) is easily determined by the resistance

to flow observed with ultrapure water and should be a constant for a given mem-
brane operating with a specified fluid and temperature.
I I 1
0 300 600
THROUGHPUT IN GALLONS
Figure 2.33: Through-put curve run at three different filtration rates,
The resistance of the cake of accumulated particulates, R,, is more com-

plicated; it is a variable which increases as filtration proceeds resulting in a pro-
gressively lower filtration rate at constant pressure. This is due to the continually
increasing thickness of the cake and its compaction under the pressurized condi-
tions of filtration. If Rm is defined as above, R, must also include the effect of
pore plugging within the membrane.
In the conventional filtration of particulates:
S
a'W"+AP) p
(11) R, =
A
where ~1’ = constant dependent on properties of the cake

w’ = is the weight of dry particulates per unit volume of filtrate
Vt = volume of filtrate delivered or “throughput”

AP = pressure drop
s = compressibility exponent of the cake. (s is zero for a per-
fectly noncompressible cake and unity for a perfectly com-
pressible cake; normal values range between 0.1 to 0.6 for
commercial slurries.)
WIicrofiltration 97
P = viscosity of the filtrate

A = area of the filtering surface
Combining equations (IO) and (I 1)
AP
J= U’ W’Vt(AF)Sp
(12)
A + %l
Thus, the flux declines as the throughput increases.
Inverting equation (I 2)
a'w'Vt (AP)~-'J.I R,
(13) -J-=
J A +ap
If one assumes that the limiting resistance to flow is that due to accumulated
particulates on the membrane or within the pores, equation (12) becomes:
(14)
J= he!2
a’w’Vtu
It is interesting to note that for a “perfectly compressible” cake, the flux

becomes independent of pressure. Higher pressures simply increase the resistance
to flow of the cake enough to offset increases in the flow rate due to the higher
driving force (API.
Rearranging equation (14) gives us the dependence of throughput on other
variables.
(15)
Equation (15) shows that as a first approximation, the throughput (total

volume processed) per square foot of membrane area will be inversely propor-
tional to the flow intensity or flux (in gal/min/ft2). Figure 2.34 shows that
equation (15) is a good approximation in at least one case. In other cases (see
Figure 2.35), the throughput seems to increase more sharply for a decrease in
flow intensity. This may be due to an inertial impaction phenomena, which
equation (15) does not take into account. Particles smaller than the rated pore
size may be captured on the membrane at high velocities, whereas at low veloci-
ties, they may follow the flow streamlines more eqsily and pass through the
membrane without capture.
The variation of throughput with filtration rate has far-reaching implications
on the most economical way to run conventional membrane filtration processes.
For example, if a fixed volumetric flow rate (gal/min) must be filtered, the total
volume (in gallons) which may be processed before plugging (run to a set API is
proportional to the square of the membrane area. [J in equation (15) has units
of gal/min/ft2] . Thus, increasing the fixed capital investment (in housings, etc.)
by two times (to increase the membrane area by a factor of 2) will increase the
volume processed (to membrane exhaustion) by a factor of four. Consequently,
replacement costs of membranes will be reduced by a factor of 2.
Incidentally, this also explains how “capillary-pore” membrane cartridges
can equal the throughput of “tortuous-pore” cartridges. Two to three times the
area of the “tortuous-pore” membranes can be pleated into a similar cartridge
because the capillary-pore membranes are so much thinner.
I I
1.0 10.0
Normalized flow rote, QoVmin
Figure 2.34: Tap-water through-put inversely proportional to filtration rate.
0.1 1.0 to.0

Flow Intensity, gol/~mlnMt*)
Figure 2.35: Dramatic decrease in through-put with increasing flow intensity.

Microfiltration 99
Backwashing
The life of conventional filter media can often be extended by “back-
washing”. This technique has also been applied with moderate success to MF
membranes.
In plate and frame systems (see Figure 2.36), the filtrate itself is often used
as the backwash fluid. Sometimes, larger membrane housings are used to serve
as an accumulation reservoir for the filtrate. A backwash fluid outlet is provided
up-stream of the membrane to purge “backwash debris” from the system. Of
course, the membrane must be supported on both sides.
As might be suspected, “capillary-pore” membranes appear to be more am-
menable to backwashing than “tortuous-pore” membranes. However, some proc-
ess streams deposit particulates on the membrane that cannot be backwashed
from either type. Figure 2.37 shows a relatively successful backwash experiment
on “capillary-pore” membranes used to filter beer.
Though not recommended by the manufacturers, pleated cartridges of both
“tortuous-pore” and “capillary-pore” membranes are now backwashed in several
wineries. Current pleated cartridge design gives the membrane good support in
the forward-flow direction but poor support in the reverse direction. Neverthe-
less, several users have extended the life of their cartridges significantly using
low-pressure backwashing.
Cross-Flow Filtration
Ultrafiltration and reverse osmosis have always used a fluid management
technique known as “cross-flow filtration” to sweep away deposited particles
from the membrane surface. “Cross-flow filtration” (CFF) is compared with
“through-flow filtration” (TFF) (sometimes called “dead-ended filtration”) in
Figure 2.38.
Currently, most MF systems operate with the more conventional “through-
flow filtration”-one stream in and one stream out. The particles accumulate on
the membrane and are disposed of with the membrane.
In “cross-flow filtration” , with one stream in and two streams out, the ob-
jective is to extend the life of the membrane indefinitely. Generally, pumping
energy must be supplied to recirculate the cross-flow stream at velocities 10 to
100 times higher than the permeation rate (see Figure 2.39). Particles removed
from the process stream are not all deposited on the filter; most are circulated as
a retentate stream with ever-increasing concentration. A bleed stream (reject
stream) may be removed to keep the concentration constant. In some applica-
tions, like cell harvesting, the product is the concentrated retentate stream. In
other applications, where the product is the filtrate, the disposal of the reten-
tate stream becomes a problem, Indeed, in some applications, having the par-
ticles collected on the filter facilitates disposal and favors the more conventional
TFF.
Figure 2.40 shows the cross-flow concentration of yeast with a “capillary-
pore” membrane. Since the yeast particles are considerably larger than the
0.2 ~1pore size, internal pore fouling was nil. The sweeping action of the cross-
flow stream tangential to the membrane surface maintained a stable flux at con-
stant concentration. The ability to concentrate yeast from 0.1 to 10% with a
stable flux would be impossible with TFF.
r_______-____‘;
I z*2 1;
I -0
I =
pi ;
:______________: 0’ E
Microfiltration 101
5: 0
m is
U!UJ/l ‘aJDJ MO1 j
1. Through Flow Filtration
Filter medium (gravel, sand,

charcoal, diatomaceous earth
t a,
FEATURES:
(i) Permeate and feed flow directions
are the same
(ii) Inherently unsteady operation
(iii) Requires frequent backwashing
II. Cross Flow Filtration

i[ ..*.;.:.:.. . . . . . .y::.s..; .,.,.;.:.,
....:.:.....::~..t’Z;:!:~,.:.~.. ........‘:..
.. .. . . ..:..:...... .* ..
FEATURES:
ii) Permeate flow direction is perpendicular
to that of the feed flow
(ii) Particle polarization is prevented by
shear induced by the feed flow
Figure 2.38: Through-flow filtration (TFF) and cross-flow filtration (CFF).
REJECT STREAM OR
PROCESS CONCENTRATE
STREAM iSay 10% of Feed)
FEED
RECIRCULATION LOOP
l
------
PUMP MEMBRANE
@w SO% of Food1
Figure 2.39: Schematic of recirculation loop for cross-flow filtration (CFF).

Microfiltration 103
Yeast concentration, vol %
Figure 2.40: Cross-flow concentration of yeast with a capillary-pore membrane.
For suspensions of particles with sizes nearer to the pore size, some internal
pore fouling will occur but at a greatly reduced rate. Figure 2.41z2 shows cross-
flow filtration of a single cell protein suspension on a “tortuous-pore” mem-
brane. The flux declines rapidly at first, as boundary layer conditions are estab-
lished, and then levels off with a diminishing rate of flux dewy.
0 RUN 24
DATA OBTAINED Al 0.83 WT. %, CELL SOLIDS
0.22 p YILLIPORE FILTER
30 YIL SlRAIQHt PATH CHANNEL
CIRCULATION RATE: IO0 ml/mln
A P = 60 eaig
0 I I I I I I I I I I I I
0 2 4 6 6 IO 12 I4 16 I8 20 22 24
TIME , hrr.
Figure 2.41: Cross-flow filtration of single cell protein suspension (0.83 wt %)

with 0.22 micron tortuous-pore membrane.
Referring to equation (13), we can examine what happens to the cake re-
sistance (R,) in cross-flow filtration by plotting the reciprocal of the flux (l/J)
versus the throughput volume (Vt). In through-flow filtration at constant pres-
sure, we would expect to see a straight line with a slope proportional to the con-
centration of particles in the feed stream. In cross-flow filtration, we obtain re-
sults like those of Figure 2.42. The intercept with the ordinate (Vt = 0) is equal
to R,/AP, the resistance of the membrane divided by the pressure drop. As the
filtrate volume begins to accumulate, the data shows a growing cake resistance.
The break in the line is due to the sweeping action of the flow tangential to the
membrane surface. There is a finite thickness of cake that is never removed, and
there is a growing resistance due to internal pore fouling, albeit at a slower rate.
Filtrate Volume V [ II -
Figure 2.42: Increase in cake resistance (reciprocal of flux) as a function of

through-put for cross-flow filtration (CFF).
The flux decay due to internal pore fouling can often be relieved with back-
flushing or chemical cleaning as in Figure 2.43, but the ultimate solution is to
develop MF membranes which have a high degree of anisotropy. For example,
UF membranes are so anisotropic that internal pore fouling is virtually elim-
inated, The skin of the membrane, with exceedingly small pores, faces the feed
stream. Since the smallest passage-way is at the pore entrance, any species gain-
ing admittance passes completely through the membrane with no opportunity
to block internal pores. The only bottleneck is at the pore entrance which leads
into an ever expanding chamber. Unfortunately, there are no MF membranes
currently on the market which have this degree of asymmetry (compare the
cross section SEM photos of Figure 2.32 with those in Chapter 3).
Microfiltration 105
Chemical Cleaning
Backflushing
Figure 2.43: Restoration of CFF flux by backflushing and chemical cleaning.
Some inorganic membranes (alumina) have been recently introduced which

are anisotropic (see Figure 2.44) but the skin is too thick. In other words, the
pore entrance resembles a long narrow passage more than an orifice. Plugging can
occur by particles bridging across the passage. However, some improvement in
flux decay has been reported for these membranes operating with cross-flow on
certain process streams.
Figure 2.44: Cross-section photomicrograph of anisotropic ceramic MF mem-

brane.
The development of a truly anisotropic MF membrane will be a major

breakthrough for cross-flow microfiltration.
MEMBRANE CONFIGURATION
The configuration of the membrane will obviously affect cost, ease of re-
placement, and efficiency of filtration. There are currently three primary config-
urations for MF membranes in industrial use:
(1) plate and frame units

(2) pleated cartridges
(3) tubular/hollow-fiber modules
Plate and Frame Units

Figure 2.45 shows the internals of a stacked-plate membrane filter housing
accommodating up to sixty 293 mm membranes with a maximum filtration area
of 33 square feet (3.0 m2).
Only “tortuous-pore” membrane discs may be used in such a stack. The rea-
son is that the polycarbonate or polyester “capillary-pore” membranes currently
available are very thin (typically 10 j.r thick) and easily pick up an electrostatic
charge. It is almost impossible to load 293 mm discs of these membranes onto
the plates. They cling to hands and wrinkles cannot be totally eliminated. It is
possible to buy (in Japan) a heat sealed “sandwich” with polyester screens on
both sides of the “capillary-pore” membrane (see Figure 2.46) which facilitates
handling. The “sandwich” is sealed around the periphery to prevent lateral leak-
age.
After loading such a stack of membranes, it is imperative that the unit be
“bubble-pointed” to establish the integrity of the stack. It is very easy to dam-
age a membrane in the process of loading, Even a pin-hole will show a bubble-
point of less than one psi. Further, every plate has gasket and “o-ring” seals
which may be out of place. A small piece of dirt or grit in an “o-ring” groove
can also create a leak which will show up as a very low bubble-point.
Plate and frame units are also used for cross-flow filtration, Some units take
sheet stock MF membranes while others work best with a preassembled mem-
brane cassette-a sandwich of two outer layers of membrane sealed to an inner
filtrate collection screen (see Figure 2.47). A cross-flow spacer is placed between
the filter packets and stacked in a plate and frame arrangement (see Figure 2.48).
In some systems, the cross-flow spacer is a screen, but the “flow-channel” spacer
shown in Figure 2.47 is less prone to fouling.
Pleated Cartridges
Stacked plate units were the first successful configuration for large scale
MF, in spite of three distinct problems:
(1) Membrane damage during loading

(2) “O-ring” gasket seal problems
(3) High labor and downtime for membrane replacement.
Microfiltration 107
Figure 2.45: Internal components of plate and frame (stacked-plate) membrane

filter holder.
Figure 2.46: Capillary-pore membrane cassette (sandwiched between two sup-

port screens).
Figure 2.47: Cross-flow membrane cassette (two layers of membrane enclosing

inner collection screen) and flow-channel spacer.
Microfiltration 109
Figure 2.48: Plate and frame cross-flow filtration unit.
It was early recognized that, if the replaceable element contained more

membrane area (preferably in a protective package), the installation time could
be greatly reduced. Unfortunately, the early “tortuous-pore” membranes could
not be pleated without loss of integrity. Therefore, the first cartridges were a
closed-end 2 inch diameter tube almost 2 feet long with one square foot of mem-
brane wrapped around the outside and sealed longitudinally. This permitted the
fabrication of a “bubble-pointable” tube. Unfortunately, the 22 inch tube con-
tained only twice the area of a single 293 mm disc.
For noncritical applications, pleated cartridges were fabricated showing less
than full bubble point. For example, Figure 2.49 is a cartridge designed for fil-
tration for deionized water. It contains five square feet of pleated 0.5 /A mem-
brane prefilter and 0.2 @ membrane. Because the pleated membranes could not
be “bubble-pointed”, a final core wrap of 0.8 /A membrane was incorporated to
give the cartridge a bubble-point of around 1O-l 2 psi. The tighter 0.2 ccpore-size
membrane could not be used for the core-wrap because of the high pressure dif-
ferential it generated across the core wrap (only 0.2 square feet in filtration area).
Microfiltration 111
In the last decade, some manufacturers have been able to produce “bubble-
pointable” (45-50 psi) pleated cartridges. These include “tortuous-pore” nylon
and polyvinylidene fluoride membranes. The 0.2 /A cartridges show quantitative
retention of Pseudomonas diminura at challenge levels of 10” to 1 013 organisms.
For reasons enumerated earlier, “capillary-pore” membranes, though pleat-
able, will not yield satisfactory bubble-points and begin to pass Pseudomonas
diminuta at challenge levels above 103-IO’ organisms even though the mem-
brane area is 2 to 3 times greater.
Most pleated cartridges are sealed inside housings with “o-rings”. There are
external and internal “o-rings” designs, both of which are vastly superior to the
old gasket seals which can easily unseat under pressure.
Some pleated cartridges are encapsulated in a disposable plastic housing so
that, apart from external hose connections, there are no seals to worry about.
Pleated cartridges are considerably more expensive than the equivalent
membrane area in 293 mm discs (2 to 4 times as much). Some users insist that
the lower replacement labor does not off-set the additional expense. However,
the trend is definitely towards pleated cartridges for through-flow filtration. The
movement to pleated cartridges has been accelerated with the discovery that
some exhausted cartridges can be reused after backwashing.
Modified pleated cartridges have also been used in a cross-flow operating
mode.” In the simplest modification, a sleeve is placed around the cartridge and
the housing modified to withdraw the retentate stream from the bottom of the
housing (see Figure 2.50). The cross-flow stream is forced into the pleats where
it moves tangential to the membrane. Even with this simple modification, dra-
matic increases in throughput were observed in the dewatering of yeast (Figure
2.51) and activated carbon (Figure 2.52). The increase in throughput was found
to be a strong function of the cross-flow velocity-as would be expected. The
data of Figure 2.51 and 2.52 were run at a recirculation/permeate ratio of 16.
i
FILTRATE
BATCH
TANK
PUMP
I
1
RETENTATE
Figure 2.50: Modified pleated cartridge and housing for cross-flow filtration
(CFF).
70
40
30
20
10
5
4
I
20 30 4050 100 200 500 1000 2000 3000
Throughput (Liters)
Figure 2.51: Increase in through-put for cross-flow filtration of 0.5% yeast with
modified 0.2 micron capillary-pore membrane pleated cartridge.
0
10 20 30 40 50 60 70 100 200
Throughput (Liters)
Figure 2.52: Increase in through-put for cross-flow filtration of 0.25% activated

carbon with modified 0.2 micron capillary-pore membrane pleated cartridge.
More recently, a line of cross-flow pleated cartridges has been introduced

with a controlled spacing for the cross-flow path between the pleats (see Figure
2.53).% This alleviates the most serious problem with the simple modifications
described in the preceding paragraph where bunching of pleats creates a non-
uniform distribution of cross-flow channels.
Microfiltration 113
Permeate
Feed Stream
Figure 2.53: Improved cross-flow pleated cartridge with controlled flow-channel
spacing (AcrofluxTM).
Tubular/Hollow-Fiber Modules
Various “tortuous-pore” membranes are also available in tubular and hol-
low-fiber modules:
MEMBRANE PORE-SIZE TUBES HOLLOW-FIBERS

O-licrons) (I.D. in mm) (I.D. in mm)
Alumina 0.2-5.0 3-15 _
Cellulose Ester 0.2 0.37-0.6 1
Polypropylene 0.2 5.5 0.6-I .8
Polysulfone 0.1-0.4 0.5- 1.o
Polyvinyl alcohol 0.4 0.4
Polyvinylidene difluoride 0.08 25.4
The polyvinylidene difluoride (PVDF) membrane is anisotropic and might

better be called an open UF membrane. In addition, the alumina membrane is
somewhat asymmetric (see Figure 2.44). None of the other membranes are
anisotropic.
All of the tubes and hollow-fibers are self-supporting except for the PVDF
membrane which is cast on the inside of a porous polypropylene tube. In most
cases, they are potted on the ends to separate the process stream from the per-
meate (see Figure 2.54). This bundle of tubes or capillaries becomes the re-
placeable module.
APPLICATIONS
Many of the applications for MF derive from the excellent retention these
membranes have for microorganisms. Indeed, the retention for bacteria or other
organisms is often superior to what may be obtained from tighter UF and RO
membranes. We may divide large-scale MF applications according to whether
they utilize through-flow filtration (TFF) or cross-flow filtration (CFF). The
former are more common.
Through-Flow Filtration Applications

Sterilization and Particle Removal (Pharmaceuticals). A great many of the
drugs and solutions produced by the pharmaceutical industry or made up in the
hospital pharmacy have to be both sterile and relatively free of particulate mat-
ter-especially if the product is to be injected into the bloodstream. For drugs
and other products that will not withstand heat, sterilizing filtration is the only
alternative. Tissue culture media, parenteral solutions, vaccines, human plasma
fractions, antibiotics, diagnostic injectables are all sterilized by membrane fil-
ters.
Even if MF has been applied by the manufacturer, the solution may need to
be filtered again by the pharmacist due to contamination during mixing or re-
constitution. A single microorganism can grow into thousands overnight. Serious
and sometimes fatal infections can result.
There is a growing clinical evidence that particulate matter in IV solutions
can also be a serious health hazard. Direct blockage of blood vessels can occur.
For example, the partial occlusion of retinal arteries can result in blind spots.
Clot formation and emboli result because of the tendency of red cells to adhere
to particles. Granulomas have resulted from inflammatory reactions where a
particle is embedded in tissue.
The U.S. Food and Drug Administration (FDA) has specified that:
Prior to filling, large volume parenteral drug products shall be filtered

through systems having a final mean porosity of no more than 0.45 ~1.Proc-
ess specifications shall indicate the maximum time during which a filtration
system may be used. Such time lengths shall preclude microbial build-up to
levels that may affect the biological quality of the large volume parenteral,
and in no case shall it exceed 8 hours.‘s
Achieving low particulate levels in the final drug or parenteral solution usu-
ally requires filtration of the constituent water. The United States Pharmacopoea
defines specifications and methods for production of water for injection. Ultra-
pure water systems in the pharmaceutical industry use reverse osmosis, ion ex-
change, and MF just as in the electronics industry (see below). Both industries
seek to produce sterile/particle free water. However, in the pharmaceutical in-
Microfiltration 115
dustry, most water produced must also be pyrogen free. This requires charge
modified MF membranes, or UF/RO.
MF is also used to remove microorganisms and particulates from air and
other gases used in the pharmaceutical industry. Some specific applications in-
clude: vent filters; filtration of compressed air used in sterilizers, filtration of air
or nitrogen used for solution transfers or at filing lines, and filtration of air or
nitrogen used in fermentors.
Again, the need for “vent-filters” has been recognized by the FDA:
All stills and tanks holding liquid requiring microbial control shall have
air vents with nonfiber-releasing sterilizable filters capable of preventing mi-
crobial contamination of the contents. Such filters shall be designed and in-
stalled so that they do not become wet. Filters shall be sterilized and in-
stalled aseptically. Tanks requiring air vents with filters include those holding
water for manufacturing or final rinsing, water for cooling the drug product
after sterilization, liquid components, and in-process solutions.25
It is clear enough why microbe retentive filters must be used as air-vents on

fermentation vessels. The reason for the FDA requirement on storage or holding
tanks is that the movement of liquids into and out of these vessels entails the
flow of air or nitrogen to maintain a pressure balance; thus the vent provides an
entryway for airborne contamination.
For gas filtration, the membrane should be treated to make it hydrophobic
if not already inherently so. Membranes made from polytetrafluoroethylene or
polypropylene are already nonwetting, and wettable polymers are treated by the
manufacturer to render them hydrophobic. If the gas filter is hydrophilic, water
condensing on the filter or entrained by the gas will wet the pores and be retained
by capillary action unless the differential pressure across the filter exceeds the
“bubble-point” pressure. In this case, the filter is “blinded” by water and the
flow is restricted considerably.
In the case of air-vents, even if the filter has been sized adequately to han-
dle the passage of air commensurate with liquid filing and withdrawal rates, the
use of hydrophilic filters has caused the implosion or collapse of more than one
storage tank, A hydrophobic filter passes air preferentially; indeed, a water in-
trusion pressure is required to force water into the pores.
Of course, the opposite phenomenon occurs with hydrophilic liquid filters
where air bubbles are present in the liquid process stream; it is called “air-bind-
ing”. Often a hydrophobic vent-filter will be used in conjunction with the main
hydrophilic membrane filter to allow escape of accumulated air without per-
mitting liquid leakage.
Naturally, any filter used to sterilize liquids or gases must be sterile itself
when installed. Again, quoting the FDA:
Solution filters shall be sterilized and installed aseptically. The integrity

of solution filters shall be verified by an appropriate test, both prior to any
large volume parenteral solution filtering operation, and at the conclusion of
such operation before the filters are discarded. If the filter assembly fails the
test at the conclusion of the filtering operation, all materials filtered through
it during that filtering operation shall be rejected. Rejected materials may be
refiltered using filters whose integrity has been verified.2s
Therefore, membrane filters which can be autoclaved or steam sterilized are

Microfiltration 117
in demand for these applications. If the membrane or cartridge cannot be auto-

claved, other chemical sterilizing agents may be used (e.g., formaldehyde).
Sterilization and Particle Removal (Beverages). MF has been used exten-
sively in the filtration of wine and beer. It is also currently used in the clarifica-
tion of cider and other juices.
The objectives for beverage filtration are to obtain a biologically stable
product with good clarity and no deposits.
Since wine and beer are fermentation products, shelf life will be limited un-
less all yeast are removed. Unless beer is kept very cold, a few surviving pedio-
cocci, lactobacilli, or wild yeasts may grow and produce off-flavors in the beer.
Though the juices are not fermentation products, they nevertheless provide the
perfect media for bacteria growth. In all cases, it is important to produce asealed
sterile product.
There are three routes to biological stability:
Heat stabilization
Injection of chemical preservatives
Sterile filtration
Hear pasteurization is an effective means of biological stabilization, but fla-

vor is affected. The bouquet of a good wine is highly susceptible to heat. The
vintner goes to great lengths to protect his wine from oxidation, and heat ac-
celerates it. In addition, elevated temperatures can result in precipitation of pro-
teins to form haze.
Pasteurization is still the most common practice in breweries in spite of
some deterioration in flavor. A number of years ago, the brewing industry fore-
cast a tremendous increase in demand for “draft-beer”, and membrane filtration
was substituted for pasteurization. However, the “draft beer craze” never ma-
terialized and many breweries reverted back to pasteurization.
Chemical stabilization has been accomplished by the addition of sorbates,
benzoates, ascorbic acid, sulfur dioxide, or diethyl pyrocarbonate (DEPC). This
method has not been universally accepted because of the public’s general aver-
sion to chemical preservatives and undesirable side affects. For example, DEPC
is very toxic until it breaks down into ethyl alcohol, carbon dioxide, and ethyl
pyrocarbonate.
The trend is toward the third approach of sterile filtration. Almost all win-
eries use it, though many breweries still prefer flash or “in-the-bottle pasteuriza-
tion”. MF not only removes organisms but improves the clarity of the product.
Juice producers are also beginning to explore membrane filtration. One large
cranberry juice maker claims ceramic MF tubes pay for themselves in a year be-
cause of reduced labor, maintenance costs and waste. It is hoped that membrane
filtration will also eliminate the need for heating the juice which changes its fla-
vor. With the long-term trend of higher energy costs, it is expected that MF will
eventually replace all pasteurization.
MF has also made its debut in the Japanese soft drink industry for the fil-
tration of liquid sugar. Figure 2.55 is a flow schematic for two parallel plate and
frame backwash systems to remove yeast from the liquid sugar. The parallel
lines permit backwashing without interruption of the process flow stream.
c----r, COMPRESSED
AIR
FEED PUMP
AUTOMATIC
BACKWASHABLE
BACKWASH FLUID + - - - MICRO- FILTRATION
SYSTEM
0 1st STAGE PREFILTER
FLOW SCHEMATIC
0 2nd STAGE PREFILTER
(LIQUID SUGAR)
0 MEMBRANE FILTER (NUCLEPORE lpm )
0 AIR ACCUMULATOR TO
STORAGE
g, AIR FILTER
ELECTRICALLY
(MEMBRANE
OPERATED
CART.
SOLENOID
0.2pm ) TANK
FG VALVES FOR AUTOMATIC BACKWASH
Figure 2.55: Flow schematic of back-wash system for MF of liquid-sugar.

Microfiltration 119
Bottlers of carbonated soft-drinks are also experimenting with MF to re-

move bacteria and particles. Greater consistency in taste and longer-lasting car-
bonation are cited as benefits. The latter is due to the removal of particulates
which act as “nucleation sites” to cause premature carbonation loss.
Particle Removal (Semiconductor Process Fluids). The fluids that come in
contact with photo masks and wafers during integrated circuit (IC) production
are often contaminated with particles which, if not eliminated, can cause defects
and reduce production yields.
The early solid-state devices were not as sophisticated as devices produced
today. A single transistor consisted of a relatively large crystal of germanium or
silicon 0.25 inch square and 0.125 inch thick to which conductors were attached.
The manufacturing processes were relatively insensitive to outside impurities.
The development of the transistor in the early fifties and the first IC’s in
1959 made possible miniaturization on a scale never dreamed possible. Today,
it is not uncommon to have over a million components per circuit on a single
wafer. Conductor widths and spacing can be less than a micron. Thus, submicron
particles can mask critical areas during etching and other key manufacturing
steps. This results in pinholing, shorts, undercutting, internal arcing, poor adhe-
sion of photoresist coalings, and other unwanted changes in current carrying
properties. According to the commonly employed “Rule of Ten”, advanced
VLSI (very large scale integrated) microcircuits with one micron geometries are
sensitive to particulate contamination greater than 0.1 ~1.
Deionized (01) Water. DI water is crucial in the microelectronics manufac-
turing process; nearly every operation in IC manufacturing ends with a DI water
rinse. Wafer-rinsing operations following wet-chemistry cleaning, stripping and
etching processes all require ultrahigh-purity process water as does process chem-
ical-bath make up. If the water is not of sufficient quality, decreased wafer yields
or subsequent device failure may result. Table 2.7% summarizes typical require-
ments for process water used in the manufacturing of high-density microelec-
tronic devices. Obviously, these specifications are a trade-off between desired
purity and practical purity and some water supplies will require more treatment
than others to achieve these specifications.
The heart of any ultrapure water system is ion exchange (IX). It was clearly
recognized that the presence of mobile ions such as sodium would produce leaky
junctions and threshold voltage shifts in the final device. No other process, in-
cluding reverse osmosis (RO), is as effective as IX in removing ions to produce
high resistivity water (IBM&m). For a typical feedwater with 200 ppm TDS
(total dissolved solids), a single pass through an RO membrane can reduce the
TDS to 2-20 ppm, and a double pass to 0.02-2 ppm. As shown in Table 2.8, the
resistivity of the permeate from a double pass RO system is still insufficient (0.1
to 10 M&m). Thus, all ultrapure water systems utilize IX usually in the form of
separate cation and anion resin beds followed by mixed bed polishing columns.
Later, it was discovered that 18 M&m water could still contain submicron
particles and organics which introduce defects into the IC. Indeed, IX seems to
aggravate the problem. The accumulation of organic debris on the resins provides
an almost ideal growth environment for bacteria which multiply at an exponen-
tial rate. They quickly foul the resins and overrun the distribution system. Fur-
ther, because of the constant agitation and tumbling of the resin beads under
forced-flow conditions, there is constant mechanical breakdown of the resins
themselves with the generation of resin fines and other particulates. Thus, there
is a need for MF to remove all bacteria and other submicron particles after de-
mineralization.
Table 2.7: Typical Requirements for Process Water Used in Microelectronics

Device Fabrication%
Parameter level
Resistivity (at 25OC) 18 Ma-cm

(90% of the time)
17 Ma-cm, minimum
SiO, (total), maximum 75 x 10-l mg/L tppm)
Particle count, two particles larger than
maximum 1 pm per milliliter
Living organisms,
maximum one colony per 100 ml
Total organic carbon,
maximum 200 x 10s3 mg/L (ppm)
Total solids, maximum 10 x lo-) mg/L (ppm)
Sodium, maximum 2 x 10-l mg/L (ppm)
Chloride, maximum 10 x lo-) mg/L (ppm)
Table 2.8: Resistivity of DI Water as a Function of T.D.S.
Approximate
Equivalent
Resistance of Extraneous
at 25’C Electrolytes
Water Purity (Ma-cm) (ppm)
Pure 0.1 or more 2 to 5
Very Pure 1.0 or more 0.2 to 0.5
Ultrapure 10 or more 0.01 to 0.02
Theoretical 18.3 0.00
A typical ultrapure water system is shown in Figure 2.56. As stated before,

the heart of the system is the IX demineralizers. All other components of the
system are present to make the IX more efficient or to alleviate problems cre-
ated by the IX columns-such as particle generation.
Microfiltration 121
E
sl
._
LL
RO with its attendant pretreatment (to prevent membrane fouling and hy-
drolysis) precedes IX to reduce the ionic load on these columns. Although IX re-
generation costs are directly proportional to the TDS, RO operating costs are
nearly independent of TDS in the normal range of operation. As shown in Fig-
ure 2.57, the combination of RO + IX is often more cost effective at TDS val-
ues below 300 ppm. Indeed some economic analysis show cross-over values as
low as 100 ppm.
R.0. x
loo 200 300 4Oll 500 600 700 I 10

T.D.S.(ppm ae CaCOJ
Figure 2.57: Total operating costs as a function of T.D.S. for ion exchange
(1.X.) with and without reverse osmosis (RO).
In any event, most ultrapure water systems utilize RO to help prevent foul-
ing of the IX columns (resulting in decreased exchange capacity). The foulants
that most often plague cation resins are hydrous oxides of iron, manganese, cop-
per, aluminum and magnesium, calcium sulfate, oil, grease, and suspended mat-
ter. Anion resin fouling is generally caused by colloidal silica and high-molecular-
weight organic acids. RO can effectively remove many of these foulants inhibit-
ing bacteria growth, and upgrading general system performance.
Since acid addition is normally used to prevent scaling in the RO unit and
to prevent hydrolysis of cellulose acetate RO membranes, the reaction of the
acid with bicarbonate in the feedwater produces carbon dioxide which can fur-
ther deplete the ion exchange capacity. Forced-draft degasification removes the
CO?.
After passage through the cation and anion exchange columns (primary de-
ionization), the water is typically placed in a storage tank which acts as a buffer
during periods of peak usage. Since high resistivity water is very corrosive and
will leach ions from tanks and piping, one of the fastest ways to degrade this wa-
ter is to allow it to stand still for any length of time. Therefore, a continuously
recirculating loop (see Figure 2.58) with a polishing mixed-bed deionizer is uti-
lized by upgrade and maintain high quality 18 M&m water. Ultraviolet steriliza-
Microfiltration 123
tion is used to kill up to 99.9% of the bacteria to keep the total population in
the system within bounds. This reduces the load on the 0.2 micron MF car-
tridges.
3rd POLlSlilNGLOOP
Figure 2.58: Flow schematic of recirculating polishing loop for D.I. water.
MF is used to remove bacteria and other particulates both in the recirculat-

ing loop and also at the point-of-use. Even though piping in the recirculating and
distribution system is made of unpigmented polyvinylidene difluoride (PVDF),
which does not leach like polyvinyl chloride (PVC) piping, particles can still
build up on pipe walls and sluff off in time. UF has also been used in the recir-
culating loop as well as at the point-of-use. When it is utilized only in the recir-
culating loop, MF cartridges are still used for insurance at the point-of-use.
Selection of the optimum MF cartridge for this application must take into
account conductivity rinse-down and particle sloughing as well as retention and
throughput.
When first installed on a 18 Mncm water line, all pleated cartridges degrade
the water to a much lower resistivity. Even though the cartridges are prerinsed in
ultrapure water by the manufacturer, after drying, initial exposure to ultrapure
water results in further leaching of ions from the various components in the car-
tridge. Depending on the type of cartridge, the time for conductivity rinse-down
or resistivity recovery of the filtrate can vary from as little as 2 minutes to as
much as 60 minutes. The water filtered by MF is expensive; a fair amount has al-
ready been invested in pretreatment by RO, IX, etc. Therefore, the rinse-down
time can have a significant economic impact on water treatment costs as well as
plant down-time, particularly if cartridges must be replaced frequently.
In addition, particle sloughing can be a problem with some cartridges. Fig-
ure 2.49 shows the support screen which is downstream of the membrane. Even
though rinsed in ultrapure water for long times, most support screens continue
to slough submicron particles even after many gallons of throughput. Thus, the
MF membrane may do an excellent job of removing particles while the support
screen is continually generating other plastic particles in the filtrate.
Chemicals. Wet etchants and cleaners, photoresist developers, dielectric
polymers, dopants and cleaning solvents are all a potential source of impurities
which may reduce semiconductor device yields. For example, the need for con-
trolling contamination in photoresist and related photolithographic chemicals
(e.g., adhesion promoters such as HMDS) is particularly critical in high density
devices. Even though the photochemical manufacturers usually filter their prod-
ucts to 0.2 p before packaging, degradation of the resists can take place due to
autopolymerization-generating “gel slugs”. In addition, insoluble particles gen-
erated from the photoresist pump at the point-of-use can recontaminate the ma-
terial as it is being dispensed on the wafer. MF membranes or cartridges at the
point-of-use ensure high production yields.
PTFE membranes or cartridges are ideally suited to aggressive chemical en-
vironments. Since these membranes are hydrophobic, they will not wet spon-
taneously with fluids having a surface tension greater than 32 dynes/cm. This in-
cludes aqueous solutions and even some organic solvents like ethylene glycol
(44 dynes/cm). In these cases, the membrane must first be wetted with a mis-
cible organic solvent such as isopropyl alcohol (27 dynes/cm). The aqueous so-
lution may then be used to displace the wetting fluid and the pores will remain
water wet. If there are air bubbles or entrained gases in the liquid, the membrane
may need to be rewetted periodically.
Gases. Inert gases are used to dry silicon wafers. Even high purity gasescan
contain more than 1,000 particles per cubic foot. In addition, particles can be
generated from moving parts in compressors, valves, and piping. Point-of-use
membrane filtration and central process gas filtration are recommended to min-
imize the particle levels in the entire system.
Particle Removal (Nuclear Power Industry). There is a build up of corrosion
products in the primary coolant loop of boiling water reactors (BWR’s) which
must be removed. These radioactive materials, primarily iron oxides (hematite,
magnetite, etc.) with small amount of copper, nickel, aluminum and zinc, are
referred to as “crud”. Previously, precoat filters have been used to remove the
crud. However, the precoat material including the RAD wastes represents a
sizeable volume of material which must be stored and treated for final disposal.
For this reason, nuclear power plants have looked at MF and UF for removal of
the crud. Both through-flow and cross-flow systems have been investigated. One
advantage of through-flow MF is that the RAD wastes are deposited on the fil-
ter which may be disposed of more conveniently. However, the replacement of
cartridge filters is a major undertaking because of the radioactivity hazard. There-
fore, high throughputs are important.
Several plate and frame MF units have been installed in Japanese nuclear
power plants. The units operate with automatic periodic backwash using ac-
cumulated filtrate. Typically, the units backwash every 3 hours at a back pres-
sure twice that of the final forward pressure. The backwashed crud is removed
from the system as concentrated RAD wastes. In this sense, these back-wash sys-
tems are not dissimilar to cross-flow units using MF or UF-membranes. In either
case, a reject stream containing concentrated crud must be disposed of.
Cross-Flow Filtration Applications

Removal of Heavy Metals. Federal, state and local regulations place strict
limits on the quantities of heavy metals which may be released to the environ-
ment. The controlled metals include: Ag, As, Cd, Cr, Cu, Pb, Hg, Ni, Sb and Zn.
U.S. electroplaters, metal finishers, and printed-circuit-board manufacturers are
under mounting pressure to clean up their waste waters.
Microfiltration 125
M F may be used to remove these heavy metals provided pretreatment chem-

icals are added to precipitate the metals to particles of filterable size. The chemi-
cal pretreatment step is crucial since it will affect the performance of the mem-
brane and the resultant sludge volume as well as the contaminant removal effi-
ciency. Reduction/oxidation, absorption/oxidation, and/or catalytic reactions
are utilized along with pH adjustment to provide the optimum precipitation. Al-
though conventional methods of waste water treatment may use a similar pre-
treatment chemistry, the final solid/liquid separation by gravity settling is usu-
ally not as effective as membrane filtration.
Typically, the cross-flow membrane modules (usually tubular) concentrate
the slurry to 2-10% solids. Further concentration to 20% solids is accomplished
by gravity settling, and if need be, a filter press can produce a dry (30 to 50%
solids) cake for disposal. To enhance settling, polymers or electrolytes are intro-
duced to flocculate and/or coagulate the colloidal suspended solids. The poly-
mers and/or electrolytes neutralize the electrostatic surface charges which repel
particles of like charge preventing coagulation.
Figure 2.5g2’ is a typical flow sheet for treating spent process water from
printed circuit board manufacturing. Usually, MF membranes are preferred over
UF because of the higher flux. However, the MF membranes are more prone to
flux decay as discussed earlier. Therefore, periodic cleaning or backflush is nec-
essary to maintain the flux at a high level. Note the cleaning tank in Figure 2.59.
If the metals are of high value, the metal precipitate may be redissolved in
concentrated acid to recover the metals in solution.2*r2”.
Figure 2.59: Flow schematic of wastewater CFF to remove precipitated heavy

metals by MF.
Industrial Laundry Wastewater. The laundry industry is a major generator

of wastewater. In the United States, industrial laundries alone account for over
10 billion gallons (38 million M3) of wastewater annually. Wastewater from all
laundry sources accounts for approximately 10% of municipal sewer discharges.
In addition to high suspended solids and BOD loading, the levels of oil and
grease, heavy metals and other organics exceed municipal discharge standards.
Until recently, the standard method for treatment of these wastes consisted
of lime coagulation and flocculation with clarification by dissolved air flotation.
Subsequently, the underflow is polished with sand or diatomaceous earth filtra-
tion. Finally, the sludge is dewatered by vacuum filtration. The effluent quality
is variable and does not always meet discharge standards.
Pilot studies in 1982 have led to the installation of cross-flow MF in some
industrial laundries. Both chemical conditioning and the addition of adsorbing
and/or absorbing agents are necessary to render the contaminants filterable. The
permeate may be recycled into the plant. Table 2.g2’ shows the nature of the
raw wastewater and the effluent from the membrane system.
Table 2.9: Cross-Flow MF of Industrial Laundry Water
Raw Recycling Memtek

Wastewater Criteria Effluent
Assays (mg/L) (mg/L) (mg/L)
BOD 1,300 30 <30

COD 5,000 100 <lOO
O&G 1.100 10 40
Suspended 1,000 none none
Solids
Pb 4.5 0.1 <O.l
Zn 3.0 0.1 <O.l
cu 1.7 0.1 <O.l
Cr 0.9 0.1 <O.l
Ni 0.3 0.1 <O.l
Fe 40 0.1 <O.l
Chloroform 3.3 0.1 <O.l
Benzene 2.5 0.1 ND*
Perchlor- 9.1 0.1 ND
ethylene
Toluene 5.2 0.1 ND
PH ‘6-9 6-9
l Nondetectable.
Plasmapheresis. The separation of plasma from whole blood by continuous

membrane filtration represents an improvement over conventional centrifugation
techniques in terms of efficiency, safety and cost. In the past, plasmapheresis
was carried out with blood donors by collecting their whole blood in plastic bags
which were then centrifuged to separate the red cells from the plasma. The su-
pernatant plasma was then decanted and the red cells returned to the donor-
enabling plasma to be drawn from the same person as frequently as three times
per week. Most of this plasma is then processed to yield purified components
such as albumin or anti-hemophilic factor (Factor VI II).
Microfiltration 127
Continuous flow plasmapheresis is a superior alternative. Whole blood is

continuously withdrawn from the donor and red cells are returned while plasma
is continuously removed. There are two major advantages:
(1) There is less chance of a mixup, since the donor’s red cells are not
removed in bags for remote processing prior to reinfusion,
(2) The time required for collection can be reduced considerably.
Continuous flow centrifuges are now available with disposable rotating

bowls. However, these devices are expensive and quite complex, requiring the
attendance of skilled personnel during operation. Cross-flow filtration (CFF)
is potentially a less expensive and simpler separation tool for continuous plas-
mapheresis. Large scale plasmapheresis of animal blood is also facilitated by
CFF.
The early attempts at membrane filtration of blood were disappointing.
Stirred-cells were used to reduce the accumulation of blood cells on the mem-
brane, but the stirring hemolyzed the red cells and the flux was low even at high
stirring rates.
In the late sixties, Blatt and co-workers at Amicon developed a thin-channel
cross-flow device for plasmapheresis. 3o In this device, red cells and plasma could
be readily separated with a 0.6 ~1 MF membrane at an acceptable flux. As shown
in Figure 2.60, the flux increases with the cross flow velocity. However, there is
a limiting velocity above which the degree of hemolysis is unacceptable. We dis-
covered that pore sizes above 0.8 ~1 occasionally leaked nonhemolyzed red cells
while pore sizes below 0.2 /J retained some of the higher molecular weight plasma
proteins (notably albumin and lsG). Therefore, pore sizes between 0.4 and 0.6 cc
were selected with 0.6 /.I preferred because of higher plasma fluxes.
50 -
Hematocrit Of 41t.
- ACD anticoagulated
at avsrag. Of 2.5 psig
CHANNEL VELOCITY IFPS) -IN 9 MIL CHANNELS
Figure 2.60: Plasmapheresis membrane flux and hemolysis as a function of

cross-flow velocity.
Later, work at Amicon for the Red Cross31 determined that for each cross-
flow velocity (fluid shear rate) there is a critical transmembrane pressure above
which significant hemolysis occurs. Apparently, the compaction of red cells on
the membrane at higher pressures results in lysis of the cells. (For a more com-
plete description of polarization effects on the membrane, see chapter 3 on
Ultrafiltration.) Membranes having smaller pore sizes could be operated at higher
transmembrane pressures before hemolysis was observed-suggesting that the red
cells are lysed through distortion into the pores. In addition, it was discovered3’
that significant flux decay occurred not only with whole blood but with cell-free
plasma. Since individual protein molecules should pass through 0.4 and 0.6 /J
pores, plugging by lipoproteins or protein aggregates is suggested.
Plasmapheresis is currently used not only in the collection of plasma by
blood banks but also therapeutically. Plasma exchange therapy (PET), removes
harmful components in a patient’s plasma by returning his cellular components
with either replacement or purified plasma. In most instances, the offending sub-
stance is an antibody whose activity is directed against the patient’s own tissues,
or the product of an antigen-antibody reaction, which accumulates to produce
so-called “immune-complex disease”. The most striking success of this modern
form of “blood-letting” has been in the treatment of autoimmune diseases such
as Good pasture’s syndrome and myasthenia gravis. Clinical improvement has
also been reported with patients suffering from disseminated breast cancer, hep-
atic coma, and a host of other diseases and immunological/metabolic disorders.
The technique has also been used to detoxify patients from drug overdoses or
poisons.
Sometimes it is preferable to purify the patient’s own plasma rather than to
replace it. This is accomplished by treating the plasma filtrate with sorbents or
further filtration. This avoids damage to the cellular components. In the meta-
bolic disease states, the abnormal solutes are generally of low molecular weight
and/or protein bound. Anion exchange resins have been used to remove bili-
rubin and bile acids. Charcoal sorbents have been used to remove aromatic
amino acids, free fatty acids, and bile acids. In other disease states, the offending
substance may also be precipitated out of the serum at reduced temperatures
(4%). For example, rheumatoid arthritis has been treated successfully by cool-
ing the plasma to precipitate and filter out these so-called “cryoprecipitates”
before returning the plasma to the patient.
Though most of the applications for plasmapheresis are medically related,
some of the larger pharmaceutical houses are beginning to process pooled ani-
mal blood in a similar way but on a larger scale. The cellular components and
the plasma are then processed separately to obtain the desired final products.
Cell Harvesting/Washing. In recent years, the use or microorganism-based
fermentations for the production of chemical products has greatly expanded. In-
terferon, insulin, novel antibiotics, cytotoxic antitumor agents, numerous en-
zymes, fine chemicals, and even fuels have been produced in carefully controlled
fermentations. A key step in the fermentation cycle is the separation of the cells
or cellular debris from the liquid phase of the fermentation broth-commonly
called “cell harvesting”.
The products of the fermentation may be either extracellular or intracellu-
lar. In the case of extracellular products, the product must be isolated from the
cells and other soluble components of the broth/growth medium. With intracellu-
Microfiltration 129
far products, the cells must first be concentrated and then ruptured (lysed) to
free the products. Subsequently, the cell debris must be separated from the sol-
uble products.
Traditionally, the cellular biomass isseparated by centrifugation. Ammonium
sulfate is then added to the supernatant to precipitate the protein product from
the media. This is followed by further centrifugation and dialysis to remove the
residual ammonium sulfate from the protein product. Cross-flow filtration (CFF)
can replace all of these steps with a significant improvement in recovery and
yield. Indeed, CFF appears to offer many advantages over conventional separa-
tion processes like centrifugation, vacuum filtration and precipitation/dialysis
for this application.
Continuous flow or batch centrifugation separates the cells according to
their density difference, but the resultant supernatants are often turbid, even
when processed at slow rates with relatively high g-forces. Up to 25% of the cells
are lost in the supernatant. CFF can usually recover more than 99% of the cells
from the broth even when the cell density is the same as that of the broth; clear
supernatants (filtrates) are obtained. The generation of potentially harmful aero-
sols in centrifugation (when harvesting pathogenic organisms) is virtually elim-
inated with CFF due to operation in a closed system with vented tanks. Batch
sizes are unlimited with CFF whereas the rotor or bowl capacity of a centrifuge
limits the volume which can be processed at one time. Further, CFF offers the
versatility of cell washing in a single continual process known as “diafiltration”
(see chapter 3 on Ultrafiltration); new buffer or solvent may be added as fil-
trate is removed. Cell washing with centrifugation requires repetitive steps of
pelleting and resuspension.
Likewise, CFF is not plagued with losses by adsorption on rotary drum vac-
uum filters (up to 15%) or by phase changes in precipitation. Further, the addi-
tion and removal of precipitants is not necessary with CFF, thereby avoiding po-
tential contamination of the product.
There is some concern that the fluid shear forces present in CFF could dam-
age fragile cells. However, viable cell recovery has in general been higher than
with centrifugation. Figure 2.61 shows data on cell viability as a function of
time while recirculating through an open-channel plate and frame CFF system.33
Mammalian cells, like those used in these experiments, are particularly fragile
and shear sensitive. Viable cell recovery was over 95% for both mouse spleen and
mammalian cells after concentrating 6 fold from an initial concentration of IO6
cells/ml.
Of the various CFF modules available, plate and frame devices which uti-
lize screen spacers between membranes are prone to accumulation of cells on
the cross-members of the screen resulting in flow blockage. “Flow-channel
spacers” like those shown in Figure 2.47 are less prone to fouling.
Hollow fiber membrane modules have minimum hold-up volumes and can
operate without blockage of the fibers if the I.d. of the fiber is substantially
larger than the maximum cell aggregate. Until recently, hollow fiber. modules
were only available with UF membranes. The advent of polysulfone MF hollow
fibers makes possible cleaning with acid and base and even autoclaving at 121” in
some instances. Since MF membranes are prone to internal pore fouling, mem-
brane cleanability with flux recovery is particularly important.
In some cases, cell harvesting can be accomplished more efficiently with
asymmetric UF membranes which are not as susceptible to internal pore foul-

ing as the symmetrical structure of MF membranes. However, when the prod-
ucts of the fermentation are retained along with the cells by the UF membrane,
M F is the only choice.
9Q- 0 MouseSpleenCells
%Original A MammalianCells
Cell 85 -
Viability
60-
75 -
I I I I I I I I I
IO 20 30 40 So to 70 a0 90
Time (minutes)
Initial cell density: OS- 1.5~10~ cells/ml; flux: 80- 100 liters/m2/hr;
concentration: d-fold; viable cell recovery: 95- 100%.
Figure 2.61: Harvested cell viability during CFF in open-channel plate and frame
system.
Figure 2.62 shows flux data for the recovery of hemoglobin from the stroma
(lysed red cells) using a 0.2 /J pore size MF membrane. Here, the intracellular
product is hemoglobin having a molecular weight of 64,000. Theoretically, a
1,000,000 molecular weight cut-off (MWCO) UF membrane (pore size of
0.08 /J) should be able to make this separation. However, in this case, pores be-
low 0.1 /J showed an unacceptable retention for hemoglobin presumably because
of trace amounts of gamma globulins (>160,000 daltons) from the plasma
which form a retentive dynamic membrane on the surface of the UF membrane
(see chapter 3-Ultrafiltration-on the formation of gel layers). However, if the
red cells are washed free of all plasma proteins (using cross-flow MF) before
lysing, UF may be used for the final separation of stroma from hemoglobin.
Indeed, even a 300,000 MWCO UF membrane was able to pass the hemoglobin
with no significant retention.
Figure 2.40, presented earlier, is an example of concentration of whole
yeast cells. Though the concentration of cells at the end of the run was only
10%. the data imply that cells may be concentrated up to the close packed
sphere density of 75 volume percent. Indeed, yeast, and E. co/i concentrations
of 60% and 37% respectively have been harvested.
Microfiltration 131
I
UEMOGLO8A' RECOV&RY %
200 K-RO# S?TROMA (WUMAN RED CELLS)

tC-/ 0 (30 M/L o/AMV.GL> (DPO 2 F/L ZER)
160 Rs.FC/RCULAT/O# RATE - //JO ==/M/N
P/n-40 PSI Poue = 18 ns
/60
1 8 f I,,,,, \ I I I
I 2 3 4 5 6 789/O 20 30
Figure 2.62: Recovery of hemoglobin from stroma using 0.2 micron MF in

cross-flow.
The ability to concentrate yeast to high levels with CFF makes possible the
recovery of beer from tank sediment (the so-called “gelgger”), which is about
6 to 12% dry solids. In some countries, the tax on beer lost in this sediment is
not refunded, so there is considerable economic incentive for recovery. Con-
ventional processes, including yeast-presses, precoated vacuum filters and cen-
trifuges, are not able to recover beer of good quality from this residue. Tubular
MF is recommended for this application since UF membranes retain some of the
proteins and other flavor components in the beer along with the yeast.
Continuous Cell Culture. The continuous separation of cells from the prod-
ucts of fermentation by membranes opens up the possibility of a “continuous
membrane fermentor” (see Figure 2.63). This concept has been explored both
for MF and UF membranes and will be discussed more thoroughly in chapter 3
on Ultrafiltration. Suffice it to say that the continuous removal of growth-inhib-
iting metabolites from a cell culture along with the products of the fermentation
can lead to significant increases in cell densities and product yield. In some cases,
the products of the fermentation may be larger than the pores of a UF mem-
brane necessitating the use of cross-flow MF.
Prefiltration for UF. In most cases, MF is too expensive for use as a prefil-
ter. However, there are some feed streams which severely foul UF membranes
and where prefiltration with cross-flow MF is cost effective. For example, in the
UF of milk or cheese whey; fat, casein fines, coagulated protein, and microor-
ganisms all cause severe membrane fouling. In the concentration of milk, the use
of tubular MF as a prefilter increased the UF flux by 100% on the average. CFF
is required because TFF (through-flow filtration) would plug the MF membrane
immediately.
STERILE
MEDIA
FEED
STERILE
CONCENTRATED
FILTRATE
Figure 2.63: Flow schematic of continuous membrane fermentor.
Membrane Distillation. In evaporating solutions, heat consumption is re-

duced by using an increased number of effects or stages, each of which must
normally be at a different pressure from the others. However, if evaporation-
condensation is reduced to the essentials only, as in “flash” evaporation, the
only requirement is solution, vapor, and condensate separated and at suitable
temperatures.
These requirements can be met by a porous hydrophobic membrane which
excludes liquid from the pores but not vapors. For example, seawater may be
separated from its distillate by a PTFE (polytetrafluorethylene) or a PP (poly-
propylene) MF membrane. If a temperature difference is maintained across the
membrane, with the vapor pressure of the seawater higher than that of the dis-
tillate, evaporation will occur at the seawater side of the membrane and vapors
will flow through the pores to the cooler surface, where they will condense (see
Figure 2.64).34 The difference in vapor pressures provides the pressure driving
force.
The process is, in effect, a multieffect evaporation process with liquid flow-
ing from pore to pore (or “stage to stage”) but with the following advantages.
The liquids involved can be at any convenient pressure higher than the vapor
pressure and less than the membrane water intrusion pressure. Since there is
complete separation between the seawater and the distillate, there is no entrain-
ment and high purity distillate is produced. Scaling will be unlikely against a
nonwettable surface. The warm seawater is fed countercurrent to the distillate
through a membrane module designed to permit only small temperature differ-
ences across the membrane (see Figure 2.65).34 By using a heat exchanger be-
Microfiltration 133
tween the distillate and the seawater, part of the evaporation heat can be re-
covered. This recovered heat can be 80% or more if low temperature differences
are maintained and lower fluxes are acceptable.
hydrophobic
microporousmembrane
Figure 2.64: Trans membrane distillation with hydrophobic MF membrane.
Feed Distillate Concentrate
Figure 2.65: Flow schematic of trans membrane distillation (TMD) with heat
exchanger (HX) for recovery of heat of evaporation.
Hollow fibers and spiral wound modules have been constructed for this ap-
plication. A module 18 inches in diameter and 5 feet long can produce up to
1,500 GPD. Plants now exist in the Florida Keys, the Bahamas, and the Cayman
Islands ranging in size from 10,000 to 60,000 GPD. They have the highest capac-
ity per unit volume for evaporation plants now in existence. It is claimed that
these plants are more cost effective than multiple-effect evaporators, but in the
author’s view, it is doubtful they can compete with reverse osmosis.
SUMMARY AND FORECAST
Cross-flow microfiltration (CFMF) is expected to grow rapidly during the

next decade and beyond. The emerging biotechnology market promises new and
unusual applications for CFMF. Recovery of dispersed catalyst particles by CFMF
is now under investigation by the chemical process industry. In addition, facili-
tated/active transport can be carried out through liquid membranes retained in
the pores of an MF membrane by capillarity (see chapter 9-Coupled Transport
Membranes). These “supported liquid membranes” are currently being investi-
gated for gas and metal ion separations.
MF as a whole is expected to grow at an average annual increase of over 10%
during the next decade to a total market of over 1.5 billion dollars.
REFERENCES
1. Grandine, J.D., U.S. Patent 4,203,848 (May 20, 1980).

2. Gore, R.W., U.S. Patent 3,962,153 (June 8, 1976).
3. Kesting, R.E., Synthetic Polymeric Membranes, McGraw-Hill, NY, 1971
pp. 104-l 12.
4. Price, P.B. and Walker, R.M., U.S. Patent 3,303,085 (Feb. 7, 1967).
5. Jaech, J.L., Vol. 9, No. 2, May 1967, pp. 319-324.
6. Porter, M.C., “Pore Coincidence through the Film”, Nuclepore internal
memo 7128176.
7. ASTM, Maximum Pore Diameter and Permeability of Rigid Porous Filters
for Laboratory Use, ANSI/ASTM E 128-61 (Reapproved 1976) Replaced
E 128-57T.
8. ASTM, Pore Size Characteristics of Membrane Filters for Use with Aero-
space Fluids, ANSI/ASTM F316-70 (Reapproved 1976) Replaced
D 2499-69.
9. ASTM, Evaluation of Air Assay Media by the Monodisperse DOP Smoke
Test, ASTM D 2986-71.
10. Wallhausser, K.H., Pharm. lnd., Vol. 36, No. 12, 1974, pp. 931-942; Vol.
37, No. 1,1975, pp. 10-17.
11. Wallhausser, K.H., Pharm. Ind., Vol. 38, No. 2A. 1976, pp. 107-I 12.
12. Howard, G., Jr. and Duberstein, R., J. of Parenteral Drug Assoc., Vol. 34,
No. 2, March-April, 1980, pp. 95-l 02.
13. Chien, S., Usami, S., Dellenback, R.J. and Bryant, C.A., Biorheology, Vol.
8,1971, pp. 35-57.
Microfiltration 135
14. Lukaszewicz, R.C., Tanny, G.B. and Meltzer, T.H., Pharm. Tech., Nov.
1978, pp. 77-83.
15. Davis, M.A., Jones, A.G. and Trindade, H., J. Nucl. Med., Vol. 15, No. 11,
Nov. 1974, pp. 923-928.
16. Hahn, R.G., Hatlen, J.B. and Kenny, G.E., Appl. Microbiology, Vol. 19,
No. 2, Feb. 1970, pp. 317-320.
17. Cuno Microfiltration Products, Zetapor Membrane product literature
9-l 5-83.
18. Pall Corp., The Pall N6s Posidyne Membrane Filter Guide product literature
1982.
19. Tolliver, D.L. and Schroeder, H.G.,Microcontamination, June/July 1983.
20. Spurny, K.R., Lodge, J.P., Jr., Frank, E.R. and Sheesley, D.C., Environ. Sci.
Technol., Vol. 3, No. 5, May 1969, pp. 453-464.
21. Cahill, T.A., Ashbaugh, L.L., Barone, J.B., Eldred, R.A., Feeney, P.J.,
Flocchini, R.J., Goodort, Shadoan, D.J. and Wolfe, G.W., APCA Journal,
Vol. 27, No. 7, July 1977, pp. 675-678.
22. Henry, J.D., Jr., Recent Developments in Separation Science, Vol. 2, ed.
N. N. Li, CRC Press, 1972, pp. 205-225.
23. Porter, M.C. and Olson, P., U.S. Patent 4,178,248, Dec. 11, 1979.
24. Gelman Sciences, Acroflux Cartridges, product literature, July 1980.
25. Federal Register, Section 212.68, 212.72 and 212.116 Filtration, June 1,
1976.
26. Couture, S.D. and Capaccio, R.S., Microcontamination, April/May 1984,
p. 45 ff.
27. Tran, Tam V., Chem. Eng. Process., March 1985, pp. 29 ff.
28. Porter, M.C., Schratter, P. and Rigopulos, P.N., Ind. Water Eng., 8, June/July
197 1, pp. 18-24.
29. Christensen, E.R. and Delwiche, J.T., Water Res., Vol. 16,1982, pp. 729-737.
30. Blatt, W.F., Agranat, W.A. and Rigopulus, P.N., U.S. Patent 3,705,100,
Dec. 5,1982.
31. Lysaght, M.J., Colton, C.K., Castino, F. and Friedman, L.T., “The Use of
Microporous Membranes to Separate Plasma from Whole Blood”, 4th
Annual Meeting of European Society for Artificial Organs, London,
England, Nov. 28-30,1977.
32. Lysaght, M.J., Colton, C.K., Friedman, L.I. and Castino, F., “Investigation
of Factors Controlling the Rate of Plasma Filtration with Microporous
Membranes”, A. I. Ch. E. 83rd Nat. Meeting, Houston, Texas, March
20-24,1977.
33. Millipore Corp., Get the Most Out of Cell Processing, product literature No.
SD 203, Sept. 1985.
34. EnkalMembrana, Trans Membrane Distillation, product literature, 1984.
3
Ultrafiltration
Mark C. Porter
INTRODUCTION
The beginnings of ultrafiltration (UF) are coincident with that of reverse os-
mosis (RO) around 1960. Despite the fact that the term “ultrafiltration” first
appeared in the colloid literature toward the end of the last century and Bechhold,
in 1906 produced collodion membranes with pore sizes below 0.01 micron, these
membranes were little more than laboratory curiosities. The hydraulic permea-
bility was low and the pores were easily plugged.
The breakthrough, which resulted in an anisotropic RO membrane in 1959,
paved the way for the first anisotropic UF membrane in 1963. Though UF mem-
branes are porous and RO membranes are not, the evolutionary development of
both occurred in parallel. Before 1960, membranes showing the retention prop-
erties of RO and of UF were available, but both had impractical filtration rates
(flux).
In the years immediately following the end of World War I I, the United States
Government became concerned about shortages in water before the end of the
century. The U.S. Department of the Interior set up the Office of Saline Water
(OSW) and committed substantial financial resources to the development of var-
ious separation processes for water desalination. A significant portion of these
funds was dedicated to the development of membranes for desalination-with
funded programs continuing for over two decades (1950-1973). The result was
one of the most promising large-scale processes for inexpensive desalination of
seawater and brackish water-reverse osmosis (RO). It is no coincidence that the
United States is a world leader, not only in RO, but also in those technologies
that are direct spin-offs-namely UF and gas separations.
The first work in RO toward water-desalting was undertaken by Prof.
Charles E. Reid’ at the University of Florida in the mid-1950’s. He discovered
that cellulose acetate (CA) is semipermeable to seawater electrolytes.* The diffi-
136
Ultrafiltration 137
culty was that the water fluxes were too low to be interesting. Reid and co-
workers attempted to cast thinner films, but found that below 6 /.I in thickness,
the films had defects which passed salt and were too fragile to handle.
Similtaneously, and unbeknownst to Reid, Sourirajan at the University of
California, Los Angeles (UCLA) obtained the same results-94% salt rejection-
but with water fluxes even lower than those reported by Reid.
Sourirajan’s partner, Sidney Loeb, began to experiment with laboratory UF
membranes (not fully asymmetric) made by Schleicher and Schuell of cellulose
acetate (CA). Loeb heated the membranes under water (annealing) to tempera-
tures between 80” to 9O”C, thereby increasing the salt rejection from 0 to 92%,
but the water flux decreased to unacceptable levels.
Loeb uncovered the work of a French investigator, Dobry3 in a literature
search. Her work provided the clue which Loeb was looking for. Dobry dissolved
acetylated CA in an aqueous solution of magnesium perchlorate, Mg(ClO&. This
casting dope was spread as a thin film on a glass plate which was then plunged
under water. The CA precipitated around the Mg(CIO& leaving a porous film of
CA. Eventually the Mg(ClO& diffused out of the pores into the water bath.
Loeb repeated Dobry’s recipe, but the membranes were too porous and had no
rejection for salt. He attempted to increase the ratio of CA to Mg(ClO& but the
casting solution viscosity was too high. The solution to the problem, suggested
by Lloyd Graham, a UCLA graduate student, was to partially replace the aque-
ous Mg(ClO,& solution with acetone, a solvent for CA.
Loeb’s standard casting solution contained CA, acetone, water, and Mg(ClO&
in the weight percentages of 22.2,66.7, 10.0 and 1.1 .4 When immersed in ice wa-
ter, a high flux membrane was obtained albeit with low salt rejection (typically
5% or less). Loeb then annealed these membranes to 80°C to yield a salt rejec-
tion of 99%. The significant breakthrough was that the annealed membrane wa-
ter-flux was 200 times greater than Sourirajan’s CA films and 5 times greater
than the annealed S & S membrane.
The reason for this breakthrough resided in the asymmetric structure of the
membrane. When the product of the water flux and the total membrane thick-
ness was calculated, the value was 666 times greater than Sourirajan’s CA films.
The most obvious explanation was that the effective membrane thickness was
much less than the total membrane thickness. Loeb postulated the existence of
a dense skin less than 1 /.I in thickness supported by a relatively porous substrate.
Thus, the substrate provided mechanical strength and the thin skin minimized
the resistance to hydraulic permeability through the membrane. For the first
time in history, it became possible to remove salt from water (95 to 98%) at
pressures of 50 to 75 atmospheres with flux values of 10 to 15 gallons of prod-
uct water per day per square foot of membrane area (GSFD).
In the early 1960’s. the techniques utilized in the fabrication of asymmetric
RO membranes were discovered to be applicable to the production of high-flux
UF membranes with pores in the range of IO to 1000 a (see Figure P.l in the
Preface). Suddenly, UF on an industrial scale became practical. Macromolecular
separation could be carried out at modest pressures (less than 6 atm.) with as-
tonishingly high filtration rates.
Prof. Alan S. Michaels, of the Massachusetts Institute of Technology and
the founder of Amicon Corporation, cast UF membranes from polyelectrolyte
complex hydrogels in 1963. A joint development program between Amicon and
Dorr-Oliver began to search for other polymers suitable for casting asymmetric
UF membranes. By 1965, the first laboratory-scale UF membranes and cells ap-
peared on the market. The ten-year period between 1965 and 1975 was a period
of intense development where chemically and thermally resistant membranes
were made from polymers like polysulfone (PS) and even polyvinylidene difluor-
ide (PVDF) in molecular weight cut-offs (MWCO) from 500 to 1,000,000. Hol-
low fibers were also developed during this decade and a whole host of module
configurations. Tubes, plate and frame units, and spiral-wound modules became
available.
MEMBRANE STRUCTURE AND FABRICATION
Figure 3.1 is a cross section of a typical asymmetric (anisotropic) UF mem-

brane. The prominent feature of these membranes is the thin skin on the sur-
face-usually 0.1 to 1 .u in thickness. This thin skin permits high hydraulic per-
meability while the more open/porous substructure (typically 125 .u in thick-
ness) provides good mechanical support. In addition, the pore configuration vir-
tually eliminates internal pore-fouling. Since the minimum pore size is at the
membrane surface, once a solute molecule gains entrance into the pore, it easily
passes through to the other side of the membrane. The solute molecule sees an
ever-widening pore channel with no restrictions or bottlenecks leading to entrap-
ment.
Figure 3.1: Cross-section photomicrograph of asymmetric UF membrane.

Ultrafiltration 139
Additional strength is sometimes provided by casting the membrane on a

spun-bonded polyethylene or polypropylene backing (see Figure 3.2),
Figure 3.2: Cross-section of UF membrane cast on spun-bonded polyethylene

backing.
As mentioned earlier, the procedure used by Loeb and Sourirajan to make

RO membranes from CA produces a very tight UF membrane (only 5% NaCI
rejection) prior to annealing. In fact, the line of demarcation between UF and
RO usually refers to the tightest U F membranes as those able to pass salts but
retain the simple sugars. Cellulose acetate and polyelectrolytes were among the
first synthetic polymers to be used for UF membranes. Today, UF membranes
are made from a wide variety of chemically and thermally stable synthetic poly-
mers including polyvinyl chloride (PVC), polyacrylonitrile (PAN), polycarbonate
(PC), aliphatic, and aromatic polyamides (PA), polyimides (PI ), polysulfone (PS),
polyarylsulfone (PAS), and polyvinylidene difluoride (PVDF). In addition, there
are inorganic UF membranes made from zirconium and aluminum oxides.
With the exception of the inorganic membranes, all of the above are solu-
tion cast or spun from a polymer solution. For the more chemically resistant
polymers like PS, PAS and PVDF, solvents like dimethylformamide (DMF), di-
methylsulfoxide (DMSO), or dimethylacetamide (DMAc) are required.
Procedure for Casting Flat-Sheet Membranes

In general, the procedure for casting anisotropic UF membranes can be
thought of as a six-step process:
(1) Preparation of the casting solution [dissolving the polymer in an

appropriate solvent(s) with or without various additives] .
(2) Metering the casting solution onto a casting belt, a rotating drum,
or a reinforcing web/fabric.
(3) immersing the freshly cast film into a nonsolvent liquid bath (nor-
mally water) until all of the polymer has precipitated/solidified and
most of the solvent has left the film.
(4) Leaching remaining solvent out of the film.
(5) Applying surfactants, wetting agents, or plasticizers (for drying) as
needed.
(6) Drying (optional) at appropriate temperature/humidity.
Step number 3 in the above sequence is responsible for the formation of an

anisotropic membrane. In Chapter 1, Strathmann describes the difference be-
tween a vapor-phase Precipitation process (Figure 1.22) as used in MF membrane
formation and the liquid-phase precipitation process (Figure 1.23) used in UF
membrane formation. In the former case, the rate limiting step is the slow diffu-
sion of Precipitant (e.g., water vapor) from the vapor phase to the polymer solu-
tion. Since this is a relatively slow process, precipitation of polymer is also slow
resulting in fairly large Pores in the membrane. In the latter case, described here,
bringing liquid water in contact with the polymer solution results in catastrophic
precipitation under supersaturated conditions.
In an earlier paper,’ Strathmann uses the analogy of crystallization from
supersaturated solutions. With increasing supersaturation, the rate of precipita-
tion is higher, the critical nucleus size decreases and the individual crystals are
smaller (more finely dispersed). Likewise here, at the outer surface of the casting
solution, in contact with liquid water, the degree of supersaturation is extremely
high and the density of nuclei is high resulting in a finely dispersed structure
which corresponds to the final membrane skin. Once the skin is formed, it be-
comes a barrier to further diffusion of water into the bulk of the casting solu-
tion. Thus, the degree of supersaturation is significantly lowered and the precipi-
tate becomes increasingly coarse. Consequently, the pore size increases as we
move from the skin into the substructure (see Figure 3.3).
Figure 3.46 is a schematic of a “drum casting machine” which carries out
steps 2-4 in the above procedure. For small batches, the casting solution can be
made up the night before the run and put on a roll mill in sealed glass jars which
are heated by infra-red lamps. The solution must then stand until all air bubbles
are eliminated. If there is appreciable solvent-loss with time, the solution trough
shown in Figure 3.4 should be closed.
Ultrafiltration 141
RATE OF PRECIPITATION
DISTANCE FROM UPPER MEMBRANE SURFACE
Figure 3.3: Scale of heterogeneity in cast membranes-variation in pore size

with rate of precipitation and-distance into membrane.
Rinse Tank
Drain Ow low
Tap Water
SCHEMATIC DIAGRAM OF A LOEE-SOURIRAJAN MEMBRANE CASTING MACHINE
Figure 3.4: Schematic of drum casting machine.

The doctor blade on the solution trough is adjusted so as to meter a coating

thickness on the drum or fabric between 100 and 500 ~1in thickness. If the vis-
cosity of the casting solution is such that it rapidly penetrates into the fabric, a
greater thickness is required. Considerable attention must be given to obtaining
the proper match between a reinforcing fabric and the viscosity of the casting
solution. If there is too little penetration into the fabric, the bond between the
membrane and the backing will be weak, resulting in membrane delamination.
On the other hand, too much penetration will result in a considerable resistance
to flow in the backing material, which is often thicker than the membrane itself.
It will be noted that the cast film is immersed in the nonsolvent (water) of
the gel tank immediately and no roller touches the surface of the film until it is
completely gelled. Indeed, the speed of the casting process is determined by the
gelation time of the membrane and the physical dimensions of the gel tank. Be-
cause the gelation time for casting UF membranes with a liquid precipitant is
much less than that for MF membranes with a gaseous precipitant, UF casting
machines have a much higher rate of output. A 60 foot long casting machine
with environmental chambers for casting MF membranes may run as slowly as
one foot per minute. A 10 foot long casting machine with gel tank for casting
UF membranes may run as fast as ten feet per minute-an order of magnitude
faster than the MF caster. Therefore, contrary to popular myth, UF membranes
are often less expensive to make than MF membranes. This cost difference is
accentuated by the fact that an MF caster must often run for an hour before any
product comes off the machine. The feedback for adjustments to the process is
greatly delayed resulting in considerable waste material before conditions in the
environmental chambers are set. With a UF caster, product is produced almost
immediately and feedback is rapid. Further, the process is much simpler than
that required to make MF membranes.
It is advisable to use 01 water in the gel tank. The author has successfully
cast many large batches of UF membranes using tap water in the quench bath.
However, this often results in membrane product which is discolored (iron ox-
ide) and variable with the seasons as the tap water quality changes. DI water with
an 18Ma resistivity is the best standard. However, it must be borne in mind that
high resistivity water will result in faster leach rates of solvent from the gelling
casting solution which may affect the pore size of the final membrane.
When producing membranes for use in the pharmaceutical industry, the use
of depyrogenated water is required. Otherwise, pyrogens will be incorporated in
the cast membrane and some will leach out in the filtratewith use. Depyrogeniza-
tion may be accomplished by filtering water from the ion exchange columns
with a 10,000 MWCO UF membrane.
Water in the gel bath must be replaced periodically or better yet, continu-
ously to minimize the buildup of solvents or other contaminants in the quench
bath, If bacteria slime or turbid water is noted, the tank should be drained and
cleaned.
Though Figure 3.4 shows the rinse tank adjacent to the gel tank with con-
tinuous rinsing of the cast/precipitated product, in practice it is a separate opera-
tion. This is necessary because of the long residence times required in the rinse
tank to remove final traces of solvents from the membrane. A separate tank may
also be used to apply wetting agents or plasticizers.
Plasticizers are necessary for some membranes, if they are to be dried, to
Ultrafiltration 143
prevent collapse of the pores during drying. Figure 3.5 shows the capillary stress
in the small pores of a UF membrane during drying.
(I)
where AP = the pressure difference across the curved interface between

the liquid and gas.
Y= the surface tension of the liquid in the pores (‘y= 72 dynes/cm

for water).
r = radius of curvature of the interface (equal to the radius of
the pore for perfectly wetting or perfectly nonwetting pores).
The pores of UF and RO membranes are small enough that the capillary
stresses at the air-water interface can exceed the yield strength of “soft” poly-
mers. Hydrophobic polymers like PVDF can be dried without irreversible dam-
age to the pores, but must be rewetted to initiate flow. “Hard” polymers with
the larger pore sizes can also be dried without ill-effect. The “soft” polymers can
be dried if the surface tension of the liquid filling the pores is reduced. This may
be accomplished by replacing the water with a low boiling organic like alcohol or
acetone which has a low surface tension and can be evaporated without creating
large capillary stresses. Alternatively, a 15% solution of glycerin or polyethylene
glycol in water or isopropanol can be used to saturate the pores prior to drying.
This not only reduces the surface tension, but the glycol remains behind after
drying-serving as a plasticizer and wetting agent in the dry membrane. This latter
method is used by most manufacturers who ship “dry“ membranes.
1POLYMER YIELD STRESS
(*p=
-+)
SUSTAINABLE
i
MEAN PORE SiZGyr-
DISTANCE FROM SKIN SURFACE
Figure 3.5: Capillary consolidation stress in a UF membrane.

Procedure for Casting Tubes

The same six steps in the procedure for casting sheet stock can be applied to
the casting of membrane on the inside of a porous tube.
Figure 3.6’ is a schematic of a tube casting operation. In this case, the tube
is lowered over a casting bob which meters the casting solution uniformly on the
inside wall of the tube. As the tube is lowered further, it is immersed in a water
bath which precipitates the membrane. All other steps are similar to those taken
with sheet stock membranes.
The challenge is to find an inexpensive porous tube which has sufficient
strength to withstand 100 psi pressure internally for long periods of time. Por-
ous polyethylene, polypropylene, and fiberglass-reinforced epoxy support tubes
have been utilized.
(A) 0)
Inner Diameter
Casting Tube
Membrane Tube
Immersion Tank
Ice Water (O-4°C)-

MEMBRANE TUBE CASTING
Figure 3.6: Schematic of tube casting operation.
Casting Variables for Tubes and Sheets

In Chapter 1, Dr. Strathmann used a ternary phase diagram (Figure 1.13) to
describe what happens when a nonsolvent like water is added to a homogeneous
Ultrafiltration 145
solution of polymer. This is a very useful tool in understanding how we can alter
the casting variables to form membranes of varying pore size and permeability.
The tool is limited because it does not provide direct information about pore
sizes; nor does it distinguish between the dense skin and the porous sublayer.
However, if we assume that a membrane having a larger porosity overall will
have a higher porosity in the skin as well as in the substructure, and that the
higher porosity is associated with higher permeability and larger pores, the
model can be very instructive.’
The overall porosity of an anisotropic membrane is determined by the poly-
mer content of the casting solution and by the relative rates at which nonsolvent
enters and nonsolvent leaves the casting solution (i.e., the precipitation path in
the phase diagram).
Figure 3.7’ is a ternary phase diagram for a castmg dope made of CA and
acetone; it shows several different paths of precipitation for porous UF mem-
branes and dense RO membranes. Path AE produces an open UF membrane
because water enters the film faster than the solvent leaves, and CA precipitates
around a larger volume of solvent which acts as the “pore-former”. Path AD
leads to a tight RO membrane because solvent leaves the film faster than water
enters. When CA precipitates, there is less volume of solvent present resulting in
a lower porosity and smaller pores.
CELLULOSE ACETATE
rate than solve
phase
SOLVENT WATER
Figure 3.7: Ternary phase diagram for formation of UF membranes from cellu-
lose acetate.
Obviously, the simplest way to adjust the overall porosity is to adjust the
polymer content of the casting solution unless the resultant viscosity is out of
bounds. If the solids are increased by direct addition to the casting solution or
by an evaporation step (moving from point A to A’ on Figure 3.7), a lower po-
rosity and tighter membrane will result.
Various methods have been used to adjust the rate of water entry and sol-
vent removal. They include the use of additives in the casting dope or in the pre-
cipitant (water) and temperature adjustments to both.
From a thermodynamic standpoint, any additive in the casting dope would
be expected to increase the rate of water entry and decrease the rate of solvent
removal-resulting in more open membranes. This is because additives in the cast-
ing dope would be expected to reduce the chemical potential of other species in
the casting dope. Reducing the chemical potential of water in the dope will in-
crease the chemical potential driving force for water entry. Likewise, reducing
the chemical potential of solvent in the casting dope will decrease the chemical
potential driving force for solvent removal.
Additives in the water bath can have the opposite effect, decreasing the rate
of water entry into the cast film because of the decreased chemical potential of
water. This produces tighter membranes with lower porosity. One may also re-
duce the chemical potential driving force for water entry by reducing the tem-
perature of the water bath. Loeb used ice water (see Figure 3.6) in the forma-
tion of RO membranes.
On the other hand, if the additive in the water bath is the same solvent used
in the casting solution, the rate of solvent removal from the cast film is also
slowed along with the rate of water entry. The result can be a reasonably homo-
geneous pore structure with fairly large pores. For example, this technique has
been used to form MF membranes from polyvinylidene difluoride (PVDF).8
The casting solution is 16 to 19% PVDF in acetone at a temperature of 5O’C.
The gelation bath is a blend of 70 to 60% acetone in water. The result is an MF
membrane with relatively large pores (over 0.1 /.J). If the acetone concentration
in the water bath is reduced below 59%, a skin begins to form and an asymmetric
UF membrane is produced.
Figure 3.8’ shows the effect of additives in the casting solution (formamide,
ZnC12, Mg(ClO& on the porosity of CA membranes. Formamide is a standard
additive for CA-acetone casting solutions; it is well known that an increase in
formamide concentration will increase membrane permeability and decrease salt
rejection.
Water was one of the earliest additives incorporated into cellulose acetate
casting solutions resulting in more open membranes. Again, this is readily ex-
plained by the ternary phase diagram shown in Figure 3.9’ The maximum wa-
ter content, without incipient precipitation of polymer, is only slightly above
15%. At this level, only a small amount of water need diffuse in before gelation
occurs. Since very little acetone has diffused out of the casting solution, polymer
begins to precipitate around a large volume of solvent resulting in a very porous
membrane.
In addition to polymer content additives, and temperature, the nature of
the solvent can have a marked influence on the transport properties of the re-
sulting membrane. Eirich et aI9 discovered that regardless of the nature of the
polymer, the highest porosity membranes were made from casting solutions uti-
lizing solvents with high solubility parameters. Eirich investigated cellulose ace-
tate, polystyrene, polyvinyl chloride, polyvinylidene difluoride, polycarbonate,
polymethyl methacrylate, polyacrylonitrile, and polyacrylonitrile/poIyvinyl
Ultrafiltration 147
CELLULOSE ACETATE
liquid phase
SOLVENT
WATER
Figure 3.8: Effect of additives on formation of UF membranes portrayed on
ternary phase diagram.
CELLULOSE ACETATE
liquid phase
4C ETONE WATER
Figure 3.9: Effect of water in casting dope on formation of UF membranes.

chloride copolymer with a whole host of solvents. The data for cellulose acetate
are plotted in Figure 3.109 Though some of the polymers investigated were not
soluble in all of the solvents shown in Figure 3.10, the trend was the same-the
water content (porosity) increased with higher solubility parameters of the sol-
vent. Tetrahydrofuran always resulted in very dense membranes while dimethyl
sulfoxide yielded very porous membranes. Of 41 polymer-solvent systems in-
vestigated, only 7 cases were out of line. Of the 7 exceptions, 5 were dimethyl-
formamide casting solutions which gave membranes of lower porosity than the
correlation would predict.
n-methyl Dimethy
Dimethyl ocetomide pyrrolidone sulfoxidf
230- ’
I”
. Ethyl formote
Propylene oxide
‘“I
10 I
0 0% 1 1 I
10 11 12
Solvent solubility porometer
Figure 3.10: CA membrane porosity (water content) as a function of the sol-

vent used in the casting dope.
Since water (the precipitant) has a solubility parameter (23.4) higher than
any of the solvents investigated, it would appear experimentally that the dis-
parity between the solubility parameters of the solvent and the precipitant is
a key factor in understanding these results.
The solubility parameter for solvents is calculated as the square root of the
cohesive energy density (the energy of vaporization divided by the molar vol-
ume). It is a measure of all cohesive forces tending to hold the molecules of the
solvent together, whether they be dispersion (London) forces, hydrogen bonding
or dipolar interactions. In general, two solvents with similar solubility parameters
are more compatible and have a greater affinity for each other. Likewise, since a
polymer’s solubility parameter cannot be calculated because of its nonvolatility,
Ultrafiltration 149
it is normally determined experimentally by arranging a list of solvents for the

polymer in order of increasing solubility parameter; the solubility parameter for
the polymer is taken to be that at the midpoint of the soluble range.
The simplest explanation for the data of Eirich (though not suggested by
Eirich or his co-workers) seems to be related to the greater compatibility of wa-
ter with solvents having higher solubility parameters. When the polymer begins
to precipitate, the residual solvent may tend to exclude water if there is too
great a disparity between the solubility parameter. The result is a dense, low po-
rosity membrane. On the other hand, smaller differences in the solubility para-
meter would tend to favor inclusion of water resulting in a membrane with
higher porosity.
Procedure for Spinning Hollow Fibers

Figure 3.11 is a photomicrograph of a typical asymmetric hollow fiber with
the skin on the inside wall. Hollow fibers for UF have the skin on the inside
whereas hollow fibers for RO are smaller and have the skin on the outside. This
is because of the high pressure required in RO; commercially available fibers can-
not withstand internal pressures up to 400 psi and above. Thus, for RO, the feed
stream is pressurized on the outside of the fiber, and the permeate flows from
the lumen of the fiber (see Figure 3.12) .Actually, considerable research effort
has been spent in developing composite hollow fibers for RO which can with-
stand internal pressures up to 900 psi because, as we shall see, flowing the feed
stream down the lumen of the fiber greatly reduces concentration polarization
effects.
Figure 3.11: Photomicrograph of asymmetric U F hollow fiber .

Hollow fiber principle
Figure 3.12: Feed stream on outside of RO hollow fiber.
The lower pressures required in UF make possible better fluid management

with pressurized feed on the inside of the fiber (see Figure 3.13). Since most UF
hollow fibers have burst pressures ranging between 60 and 100 psi, inlet operat-
ing pressures are limited to an inlet pressure of 30 psi. Naturally, with feed in the
bore of the fiber, the inside diameter must be larger to prevent plugging. Figure
3.14 shows UF fibers made by Romicon with inside fiber diameters up to 1 mm
(1,000 ~1. This is 25 times larger than the inside diameter of a DuPont RO hol-
low fiber (42 JL I.D.). Actually, the trend is to go with even larger diameters (1.5
and 2.0 mm I.D.) to improve fouling resistance.
Figure 3.13: Feed stream on inside of UF hollow fiber.

Ultrafiltration 151
Figure 3.14: Dimensions of various hollow fibers supplied by Romicon (UF)

and by DuPont (RO).
The spinning of asymmetric hollow fibers with the skin on the inside closely
resembles the procedure used in casting flat-sheet membranes. Figure 3.151° is a
schematic diagram of a spinneret used to spin these fibers. The degassed and fil-
tered polymer solution is forced under pressure into a coaxial tube spinneret.
The liquid is extruded through an annular orifice and the hollow fiber (still liq-
uid) is stabilized and precipitated by an internal coagulating fluid (usually wa-
ter) which flows out the center tube.
Figure 3.15: Schematic of tube-in-orifice spinneret.
The spinneret is usually positioned some distance above a water bath, so

that the outside of the emerging fiber is in contact with air before being col-
lected in the water bath. If the fiber hits the water bath before being fully
gelled, a secondary skin will form on the outside of the fiber (see Figure 3.76).
Figure 3.16: Photomicrograph showing secondary skin on outside of UF hollow

fiber.
This is disadvantageous when potting the hollow fibers in a header. The potting
material (usually epoxy) will not penetrate into the wall of the fiber resulting in
a poor bond and a possible leak path for the feed stream through the cross sec-
tion of the fiber wall and out into the permeate, by-passing the internal skin.
The spinning procedure accomplishes each of the six steps used in casting
flat sheets or tubes. The spinneret design (the diameter of the center tube and
the width of the annular space between the inner tube and the extrusion orifice)
will determine the dimensions of the final fiber. Collection of the gelled fiber in
a water bath permits leaching of remaining solvent from the fiber. During long
runs, this water bath is continuously replenished with fresh water. If need be,
the fibers may b~ collected and immersed in other baths to add plasticizers, etc.
to prepare for drying before potting.
Spinning Variables for Hollow Fibers

All of the variables discussed in conjunction with casting sheet and tubes
apply to hollow fibers as well. Higher polymer content in the casting dope, ad-
ditives in the coagulating fluid, and reduced temperature of the coagulant will
result in tighter, more retentive hollow fibers. On the other hand, additives in
the casting dope and solvents with high solubility parameters favor more open
fibers with a higher permeability.
In addition to the above, there are other variables unique to the spinning
process which offer additional handles for controlling fiber permeability and re-
tentivity.
Ultrafiltration 153
(1) The velocity of the spinning solution through the annular orifice
of the spinneret.
(2) The velocity of the coagulating fluid through the inner tube.
(3) The distance between the spinneret and the collecting bath (the
“air-gap”).
In general, an increase in the spinning solution velocity and the “air-gap”

will result in tighter fibers with lower permeability and increased retentivity. An
increase in the coagulating fluid velocity will have the opposite effect, leading
to more open fibers of higher permeability and reduced retentivity.
The effect of increased spinning solution velocity in producing tighter fibers
is explained by the shear forces in the annular orifice. It is known that polymer
molecules under shear flow tend to align themselves in the direction of flow.
Since the internal wall of the fiber precipitates immediately upon contact with
the coagulant after emerging from the spinneret, this alignment of polymer
chains will be frozen in place in the solid skin. As a consequence, the pores in
the skin (which control retention and permeability) will be elongated. If the
kinetics of gelation remain unchanged, rectangular pores of the same area as
circular pores will show retention for smaller molecules and a lower permeabil-
ity. Indeed, if the width of the rectangular pore is equal to the radius of the
circular pore of the same area, the permeability through the rectangular pores
will be only 2/3 of that through the circular pores.
Likewise, an increase in the “air-gap” (distance between the spinneret and
the collecting bath), will further extend the polymer chains as they passthrough
the spinneret because of the added weight of the fiber below. Any tendency to-
wards relaxation of the polymer chains after emerging from the spinneret and
before gelation will be alleviated by this weight tending to stretch, the fiber.
Thus, an increase in the “air-gap” will only increase the degree of orientation re-
sulting in tighter fibers. However, very small “air-gaps” can result in the forma-
tion of a secondary skin on the outer surface of the fiber which may tend to
decrease the hydraulic permeability as well-though the retentivity of the fiber
(determined by the internal skin) would be more open.
The alignment of polymer chains in shear flow through the spinneret may
explain why hollow fibers generally have lower permeabilities than flat-sheet
membranes with the same retention rating.
On the other hand, an increase in the velocity of the coagulating fluid
through the inner tube tends to counteract the elongation of pores in the skin;
the increased pressure in the fiber lumen tendstoexpand itsdiameter and thereby
enlarge the width of the elongated pores. The result is a more open fiber with in-
creased permeability and decreased solute rejection.
Preparation of inorganic Membranes

Dynamic Membranes. In the late sixties, workers” at the Oak Ridge Na-
tional Laboratory discovered that cross-flow filtration of dilute colloids through
a microporous tube made the support tube retentive for macromolecules and
even salts (in some cases). While numerous colloidal materials and organic poly-
electrolytes were tested, the most successful system was based on hydrous zir-
conium oxide. This inorganic dynamic membrane could have flux values as high
as 1,000 GSFD (gallons per square foot per day) at 950 psi with a NaCl rejec-
tion of 50% at dilute salt concentrations (0.05 M). Further, the rejection could
be improved by applying a thin layer of polyacrylic acid, (PAA) at high pH on
top of the Zr(lV) oxide dynamic membrane. Rejections of 90% for 0.05 M NaCl
were obtained.12
One company (Carre Inc., Seneca, SC) has exploited this technology by sell-
ing plants utilizing porous stainless steel tubes and the know-how to apply the
hydrous Zr(lV) oxide dynamic membrane. When the membrane becomes fouled,
it is stripped off with acid and a new membrane applied using cross-flow of the
Zr( IV) oxide slurry feed.
Union Carbide went one step further-sintering the hydrous Zr(lV) oxide
dynamic membrane in place on a porous carbon tube at elevated temperatures.
This resulted in a high-temperature (300°F) inorganic membrane which is resis-
tant to a pH of I-14. Further, the membrane may be cleaned with organicsol-
vents that would dissolve conventional polymer membranes. This technology
was subsequently licensed to SFEC (Soci&i de Fabrication d’El&ments Cata-
lytiques) in Boll&e, France and to Gaston County Filtration Systems in Stanley,
NC.
Ceramic Membranes. Alumina membranes for UF have been introduced by
Ceraver (Tarbes, France) (membrane division recently purchased by Alcoa). The
Norton Co. (Worcester, MA) is reportedly about to introduce an alumina UF
membrane in addition to its current MF membrane.
The difficulty with existing ceramic UF membranes is resistance to high pH.
The alpha form of A120J is used to make MF membranes (0.2 /.r pore size and
above) and is resistant to a pH of I-14. To make the finer pores (down to 40 A)
in a UF membrane, the existing technology uses the gamma form of A1203 which
will not withstand high pH.
The methods used to make these membranes are diverse and highly propri-
etary. However, one method for preparing gamma-alumina membrane films on
porous supports (which can be made from alpha-alumina) has been reported by
Dutch researchers.13 They use a sol/gel technique with Boehmite (y-AIO-OH)
as the precursor because it can be easily dispersed with acids.
>8O=‘C
[ C&CH tCH$O$ Al + y-AIO-OH

Ws&r
Boehmi te
(preclpl tate)
ACM ) 90-c
ealciMtlal
FiJ Thermal
pore size W
Treatment t-10 mlcma
Y A& t Y A& + y -AIO-OH

film (Temp/tlme) ) 4OO’C (sol)
Ultrafiltration 155
The pore size of the membrane is determined by the type of acid and its
concentration during dispersion (typically 0.03 to 0.15 mol acid per mol of
Al&OH) along with the final thermal treatment, both of which affect the crys-
tallite size of the gamma-alumina. The porous support is dipped into the sol and
capillary action pulls tlie sol into the pores increasing the concentration of
Boehmite at the entrance of the pores to form a gel. The calcination of this gel
film (above 4OO’C) yields the final film of gamma-alumina.
PORE SIZE DETERMINATION
As shown in Figure P.2 (in the Preface), UF membrane pore sizes range
from 10 to 1000 A (0.001 to 0.1 II). It is difficult to measure the pores directly
by any of the techniques used for MF membranes. The anisotropic structure and
the wide distribution of pore sizes make this almost impossible.
Instead, we are accustomed to think of the molecular weight of the macro-
molecules which are retained by or pass through the membrane. Unfortunately,
existing UF membranes do not have a “sharp cut-off” because of the wide dis-
tribution of pores in the skin of the membrane (see Figure 3.17). Most UF mem-
branes have “diffuse cut-off” characteristics.
1.0 -
Retention
0’
1000 10,000 100,000
Log molecular wt.
Figure 3.17: Sharp vs diffuse cut-off membranes.
Molecular Weight Cut-Off and Pore Size

What we mean by the molecular weight cut-off of a “diffuse cut-off” mem-
brane must be clearly defined. The convention established by AMICON and
adopted by most UF membrane manufacturers is based on the retention of glo-
bular proteins (spherical macromolecules). The retention R (in percent) may be
defined as follows:
cuf
where
(2)
C”f =
R=lOO ( )
l - CR
concentration of the solute in the ultrafiltrate
CR = concentration of the solute in the retentate

The convention states that the molecular weight cut-off of the membrane is
equal to the molecular weight of globular proteins which are 90% retained by
the membrane.
Figure 3.16 shows how this is determined. The retention values of a series of
globular proteins (spherical molecules) are measured on the same membrane.
The molecular weight at which the retention curve crosses a retentivity of 90%
is the “molecular weight cut-off” of the membrane. This means that larger mole-
cules are said to be retained by the membrane and smaller molecules are said to
pass.
Normally UF membranes are designated by prefix letters which refer to the
membrane type followed by one or more digits which refer to the molecular
weight cut-off in thousands of daltons. For example, in Figure 3.18, UM, PM
and XM refer to three different polymers; PM-10 indicates the membrane has a
10,000 molecular weight cut-off (MWCOI. The approximate molecular diameter
@I at the top of Figure 3.18 is taken from radius of gyration data reported in
the literature. This may be used to estimate the effective pore size (see Figure
3.19).
APPROXIMATE MOLECULAR DIAMETER ifi1
IO IS 20 30 40 so 60 SO 100 Isa
X.NOMINAL MOLECULAR WEIGHT EXCLUSION OF MEMBRANE

01 * I * I . I .
I02 2 5 IO3 2 CL I04 2 5 IO’ 2 5
MOLECULAR WEIGHT
Figure 3.18: Retentivity of a series of globular proteins on various UF mem-

branes.
Id Id
MOLLCLMAR WEIOHI CUT-OFF (MWCO)
Figure 3.19: Pore size variation with MWCO.

Ultrafiltration 157
Retention of Spherical and Linear Molecules

Two different molecules of the same molecular weight can have different
configurations such that their molecular diameter is different. For example, a
linear polymer may snake through the pores while a spherical polymer of the
same molecular weight, but of greater diameter, is retained (see Figure 3.20).
Figure 3.20: Retention of spherical and linear molecules.
Table 3.114 shows the effect of size and shape of the molecule on its re-
tentivity by various MWCO membranes. Each membrane retains those mole-
cules positioned above the line but passes those below the line. Since character-
ization of a membrane’s MWCO is based on globular proteins, the retention of
these molecules is as expected. However, the retention of branched polysaccha-
rides and linear flexible proteins is lower than might be expected. For example,
Dextran 250 with a molecular weight of 236,000 passes through a 50,000
MWCO membrane. Apparently, the fluid shear in the vicinity of the pores is
high enough to uncoil free draining chains. This is illustrated in Figure 3.21
which shows the difference in the retentivity of a 1,000,000 MWCO membrane
for a series of globular proteins and a comparable series of polysaccharides.
Table 3.1: Effect of Size and Shape of Molecules on UF Retentivity

Solute C1x.s
Globular Branched Linear. Flexible

Membrane* Proteins Polysaccharides Polymers
T-Globulin (160,000)”
Diaflo XMSO Albumin (69.000)
Diiflo PM30 Pepsin (3S.000) Dextran 250 (236,000)
Cvtochrorne C (13.000) Polyacrylic acid
(pH 10; 50,000)
Insulin (5,700)
Dextran 1 IO (100,000) Polyacrylic acid
(pH 7; 50,000)
Diaflo PM10
Bacitracin (1,400)
Dextran 40 (40,000)
Dextran 10 (10.000) Polyethylene
glyc01(20,000)
Diaflo UMIO
* Molecules above J horizontal membmne line are completely retained by the membrane;
below the line partial retention or complete clearance is observed.
** Number in parentheses denote molecular weights.
looo MEMBRANE
0.1- MOLECULAS WElbW

0 4 5 6789, 2
1,06o,oac,
Figure 3.21: Retentivity of proteins and polysaccharides on a 1,OOO,OOOMWCO

membrane.
It is also known that the pH and ionic strength of solutions of polyelec-

trolytes can have a marked influence on their retentivities. The more a poly-
electrolyte is charged in solution, and the lower the ionic strength of the med-
ium, the larger the effective size of the polyelectrolyte for a given molecular
weight.
Maximum Pore Size and Effective Pore Size

Since the effective pore size is estimated from the molecular diameter of
globular proteins which are retained 90% by the membrane, it is obvious that
larger pores do exist. The measurement of a membrane’s bubble point (see the
section on the bubble point test in Chapter 2) permits calculation of the maxi-
mum pore size in the skin of the membrane.
Figure 3.22 shows the bubble point measured with isopropanol (IPA) on
polyvinylidene difluoride UF membranes with MWCO’s between 10,000 and
1 ,OOO,OOO.A 300,000 MWCO membrane (F300) should have an estimated effec-
tive pore size of 0.02 /_I;yet the bubble point indicates a maximum pore size in
the skin over 0.4 /.L This is one reason why UF membranes can be less retentive
for bacteria than MF membranes. However, Figure 3.22 also indicates that a
10,000 MWCO membrane can have a (I.P.A.) bubble point of 100 psi. Equation
3 of Chapter 2 may be used to calculate a maximum pore diameter of 0.12 /.r
which should be retentive for all bacteria. Indeed, small laboratory discs of these
membranes can be subjected to high challenge levels of bacteria with absolute re-
tention (zero passage). However, industrial scale UF modules often employ 10 to
100 square feet of membrane area; it is difficult to manufacture a pinhole-free
module with this much area. Broken fibers, bubbles in glue-line seals, and other
defects provide leak paths for bacteria.
Ultrafiltration 159
100 -f ‘El BUBSE POINT OF R _
!302# I 0
SERIES
hEH6RRNR
C4= ULTRRFILTRRTIM
(ISOPROPYLALC4HOL)
1
FIO
I _
Q-5 Fzo
‘lo-
8 \ i
f36 T\
_
/
-1
- tz p&T FYY) t_9++
-3
2.0- 4 F= F5ao
.h
EST1 M n= %RE s\%bE (RtlCROh,+-)
I I I lllll I I I IllIl_
$0, lMo2 O.Wi 6.01 0.62 403 0.0s C
Figure 3.22: IPA bubble point of a series of UF membranes.
RETENTION CHARACTERISTICS
The polymer from which a UF membrane is made does not generally affect
the retention characteristics of the membrane. However, the nature of the poly-
mer can affect adsorption of species onto the membrane surface. For example,
when processing small volumes of dilute solutions, strong adsorption of the so-
lute can diminish its concentration in both the retentate and in the filtrate, and
this can affect the apparent retention calculated from Equation 2.
Adsorption Losses
The retention of membranes are often measured in stirred cells. A mass bal-
ance on the cell; integrated over time from an initial retentate volume (V,,) and
initial concentration (Co) to a final retentate volume (Vf) and final retentate con-
centration (cf) yields the following expression for the retention (R):
ln ( cr / C,)
(3) R = loo ldvo/vr 1
Equation 3 is used to calculate retention solely on the basis of changes in

the retentate volume and concentration. Obviously, if adsorption losses are ap
preciable, the retention calculated from Equation 3 will be too low since Cf will
be lower than it would be without adsorption. For this reason, the retention
should also be calculated with reference to the solute concentration in the ultra-
filtrate (permeate).
A simple cumulative mass balance on the cell leads to another expression for
retention (R) based on the final (average) ultra filtrate concentration (Cut), the
initial concentration of the retentate (C,) and charges in the retentate volume
from V, to Vf:
R=
Again, if there are significant adsorption losses, the retention calculated

from Equation 4 will be too high since C,f will be lower than it would be with-
out adsorption.
If the initial and final retentate concentration (C, and Cf) are measured
along with the final ultrafiltrate concentration (Cur), the retentivities calcu-
lated from Equations 3 and 4 may be compared.
If there is an appreciable difference, adsorption losses can be significant.
The magnitude of the loss may be calculated from the amount of solute in the
retentate (CfVf) and in the filtrate K&D/o-Vfl 1 compared with that charged to
the cell (CoV,,).
If a small volume of a dilute solution is placed in contact with a large mem-
brane area, the adsorption from solution can be appreciable. This technique may
be used to measure the relative adsorption of various membranes as in Table 3.2,
Fifty ml of a dilute solution (150 micrograms/ml) of cytochrome “C” were con-
centrated to 10 ml in a 50 ml stirred cell (membrane area of 13.4 cm’) to com-
pare the various membranes listed in Table 3.2.
Table 3.2: Adsorption Lossesof Cytochrome C on Various UF Membranes
Membrane Types Adsorption Retention for

loss(%) CytochromeC(W1
Nudepore C- IO k4MW8t9d c8lIulessAceWe 0.8 97
AmimnYM-10 @JfW8td &hJloSe Acetate 2.3 97
AmimnUN- 10 Polyelactrolyte
Complex 4.3 26
MilliporePTOC
( I 04) Polysulfons 11.3 a99
Nuclepore
A- IO Polyamide 12.4 85
Nuclepore
F- IO Polyvinylkbnedifluerids 22.5 >99
AmlmnPM-IO Polyarylsulfens 24.6 53
In general, hydrophilic polymers tend to have a lower adsorption than the

hydrophobic polymers.
Ultrafiltration 161
“Charged” Membranes
Polyelectrolyte complex UF membranes can be made with a net charge to
obtain moderate rejection of salts or amino acids (75 to 90%) at relatively high
flux and low pressure.‘St16,For example, the Millipore PSAL membranes contain
fixed sulfonic acid groups which have a negative charge capacity of 800 milli-
mols/liter.
Charged UF membranes reject low concentrations of salts primarily by the
Donnan exclusion mechanism. Because the fixed charged groups on the mem-
brane skin reject ionic solutes via repulsion of coions, the rejection would be ex-
pected to depend on solute type and coion charge. Obviously, divalent and tri-
valent ions are rejected better than monovalent ions. Highly hydrated ions are
rejected better than poorly hydrated ions.
Effect of Pressure
Figure 3.2317 presents rejection data for three different dextrans on the
same UF membrane as a function of pressure. Figure 3.24ls shows the effect of
pressure on the rejection of 0.1 M NaCl by a cellulose acetate RO membrane. It
is obvious that the mechanism for solute transport through the membrane is dif-
ferent for UF and RO.
In the case of RO (Figure 3.24), the increase in salt rejection with increasing
pressure suggests a “solution-diffusion” model as opposed to a “pore-flow”
model for transport of solvent and solute across the membrane. It is known that
although the water flux through RO membranes increases with pressure, the salt
flux is almost invariant with pressure. The net result is that a higher water flux
dilutes the salt concentration in the filtrate resulting in a higher calculated rejec-
tion by Equation 2. If “pore-flow” were in effect, leakage of salt through large
pores would also be expected to increase with pressure. This suggests that the
RO membrane acts as a nonporous diffusion barrier.
FIGURES IN BRACKETS INDICATE ORDER OF RUNS
1 I I I
20 40 60 80
TRANSMEMBRANE PRESSURE DROP, psi
Figure 3.23: Effect of pressure on UF rejection of dextrans.

100
90 -
l 0 increasing
60 - pressure
l Decreasing
pressure
50 I I I I I I I I
0 10 20 30 40 50 60 70 80 90
Applied pressure, atm
Figure 3.24: Effect of pressure on RO rejection of 0.1 M NaCi.
in a “solution-diffusion” process, the rate of transport of water and salt is

proportional to the chemical potential gradient of each species across the mem-
brane. The chemical potential of each is affected by pressure and concentration.
For water, there is little difference in concentration across the membrane, the
dominant effect is the difference in pressure. For salt, the ratio of upstream to
downstream salt concentration is very large and this dominates over the effect
of the pressure difference. The water flux is higher than the salt flux because the
soiubiiity of water in the membrane and its diffusivity through the membrane is
higher than that of salt.
The “solution-diffusion” model explains why molecules larger than salt
sometimes pass through an RO membrane more readily. For example, cellulose
acetate membranes which show a 95% rejection for NaCi (MW 58) and a 99% re-
jection for dextrose (MW 180). show a negative -34% rejection for 2,4-dichloro-
phenol (MW 163). This means the dichiorophenoi passesthrough the membrane
more readily than water. This is hard to explain with a “pore-flow” sieving
mechanism.
in a “pore-flow process”, the solute flux (J,) should be proportional to the
water flux (J,) which carries solute to the membrane and through the larger
pores.
Ultrafiltration 163
(5) Js = u J&s
where Cs = concentration of solute at the upstream surface of the
membrane
c = the fraction of the total liquid flowing through pores large
enough to pass solute molecules
The retention for solute may be expressed as:
JS
(6) R= I- J
W
or
(7) R= 1 - UC,
Thus, in a “pore-flow” model, R would not be expected to increase with

pressure since the solute and solvent flux are “coupled”. As the pressure is in-
creased, both fluxes increase.
However, the phenomenon of “concentration polarization” may cause the
rejection to decrease with increasing pressure as in Figure 3.23. The reason is
that the concentration of solute at the surface of the membrane (C,) has increased
over that in the bulk stream (Cc), As the pressure is increased, the solvent flux is
increased and the convective transport of solute (J,C) to the membrane is in-
creased (see Figure 3.25). If the solute is retained by the membrane, it accumu-
lates at the surface of the membrane until the back diffusive mass transport,
D,(dc/dx), is equal to the forward convective transport. Even if a steady state is
maintained, there must be a concentration gradient (dc/dx) to remove solute
from the membrane. Thus, Cs will increase with pressure resulting in a decrease
in solute retention by Equation 7 and as shown in Figure 3.23. The pores large
enough to pass solute, at higher pressures, pass more solvent, carrying a higher
concentration of solute (Cs).
‘b
J, (=J,C)
Figure 3.25: Concentration polarization in UF.

Further increases in pressure will increase the solute concentration at the

surface of the membrane (C,) to a limiting concentration. Proteins begin to form
a semisolid gel on the membrane surface with a gel concentration (C,) between
20 and 50 wt %. Colloidal suspensions form a densely packed layer of close-
packed spheres usually between 70 and 80 volume %. Under these conditions
(see Figure 3.26), the membrane is said to be “gel-polarized“, and further in-
creases in pressure will not increase C,. Therefore, once the membrane is “gel-
polarized”, the retention should be independent of pressure (see Equation 7). In
Figure 3.23, the rejection of Dextran-80 is beginning to level out at the higher
pressures as the membrane becomes gel-polarized.
MEMBRANE
LAMINAR \
SUBLAYER “CAKE” (SLIME)
(NON NEih’TONIAN)
Figure 3.26: Gel formation due to concentration polarization.
Fractionation of Solutes
The retention of a single solute is determined solely by the properties of
the membrane (pore size distribution, charge, and adsorption characteristics) and
by the operating conditions (e.g., pressure, pH and ionic strength of the solu-
tion). The retention of individual solutes from a mixture is more complicated.
Fractionation may or may not be possible.
Consider a 100,000 MWCO membrane which has a retention for gamma glo-
bulin (160,000 daltons) of 95% and a retention for albumin (67,000 daltons) of
less than 15%. The retention for albumin in the presence of gamma globulin is
shown in Figure 3.27. Because the gamma globulin is retained by the membrane,
it forms a gel-layer which is retentive for albumin. Without this “dynamic sec-
ondary membrane”, the albumin would pass the primary membrane. Indeed, at
low concentrations of gamma globulin, the rejection of albumin approaches the
single solute rejection by the membrane.
Ultrafiltration 165
Figure 3.27: Retention of albumin in the presence of gamma globulin.
The pharmaceutical industry would prefer to make the gamma globulin/al-

bumin separation with a UF membrane rather than Cohn fractionation (sequen-
tial precipitation with ethanol). But to do so, they would have to dilute the mix-
ture way down to accomplish the separation. The processing of the large diluted
volumes followed by concentration of the two fractions would make the mem-
brane process more cumbersome and expensive than Cohn fractionation.
Separations involving globular proteins appear to be the most difficult. Fig-
ure 3.28 shows how three different solutes which are retained by a 30,000 MWCO
membrane influence the retention of a 0.2% solution of cytochrome C (12,400
daltons) which by itself passes through the membrane with 0% retention.
200 I
0 A OVALEUMIN (45,000 MW)
W 0 CHYMOTRYPSINOGEN (26,000 MW)
8 0 ALBUMIN (68,000 MW)
I 100
g
K
0
LL
o 50-
z
F:
3 30-
t;
ez,
l- SYSTEM: l/2X UF
wz 0.2% CYTOCHROME C
PM-30 MEMBRANE
$
STIRRED CELL (5-55 psi)
ti
IO ’ I
0.1 0.3 1.0 IO
CONCENTRATION OF MACROSOLUT:
Figure 3.28: Retention of cytochrome C in the presence of larger species.
On the other hand, polymers like hydroxyethyl starch, polyethylene glycol,

and polyvinyl pyrrolidone (see Figure 3.29)” can be fractionated more readily.
MOLECULAR WEIGHT DISTRIBUTION OF DIAFILTERED

POLYVINYLPYRROLIDONE
DIAFLO XM-48 MEMBRANE
0.20 I I I I I
RElENTAlE mi 2ooJJool
Figure 3.29: Fractionation of PVP.
MEMBRANE FLUX WITH CONCENTRATION POLARIZATION
The effect of “concentration polarization” on retention was discussed in the

previous section. In this section, it will be seen that concentration polarization
can severely limit the flux. The control of polarization by proper fluid manage-
ment techniques is essential to the economic feasibility of the process.
Without the development of an anisotropic UF membrane, UF would not be
a commercial process today. The thin skin minimizes the resistance to flow, and
the asymmetry of the pores virtually eliminates internal pore fouling. However,
the hydraulic permeability of these membranes also increases the convective
ransport of solutes to the membrane surface. Consequently, the polarization
modulus (defined as the ratio of the solute concentration at the membrane sur-
face, C,, to that in the bulk process stream, Cb) is higher than that experienced
with lower permeability RO membranes.
This accumulation of solute at the membrane interface (see Figure 3.25 can
severely limit the flux. In the case of RO, the salts retained have a significant os-
motic pressure (a) and the effective pressure gradient is reduced by the osmotic
pressure difference across the membrane (An), thereby reducing the flux. Some
researchers have used this “osmotic pressure model” in an attempt to explain
the effect of concentration polarization on UF membrane flux.20-22 Even though
the macromolecules and colloidal suspensions retained by UF membranes are
quite large and have negligible osmotic pressures, it is argued that the high solute
Ultrafiltration 167
concentrations at the membrane surface can result in an osmotic pressure dif-

ference across the membrane which should be taken into account. At present,
the “gel-polarization model” appears to do a better job predicting the UF flux
for a wider range of process streams than the “osmotic-pressure model”. There-
fore, the following treatment will neglect the osmotic pressure.
Gel-Polarization
If the transmembrane pressure drop (Ap) and the solute concentration in
the bulk process stream (Cb) are high enough, the concentration at the mem-
brane surface (C,) can rise to the point of incipient gel precipitation forming a
dynamic “secondary membrane” on top of the primary structure (see Figure
3.26). This “secondary membrane” can offer the major resistance to flow. In a
stagnant “dead-ended” system, the gel layer will grow in thickness until the pres-
sure activated convective transport of solute with solvent towards the membrane
surface just equals the concentration gradient activated diffusive transport away
from the surface. Thus, the flux in stagnant “dead-ended” systems is often so
small as to be virtually nonexistent unless the bulk stream concentration is ex-
tremely low. Furthermore, by the very nature of the process, increased pressures
will not help since the gel layer only grows thicker to offer more resistance to
the increased driving force.
(8) Jw=_ ‘1
where Jw = water flux (volume/time/membrane area)

AP = transmembrane pressure drop
Rm = hydraulic resistance of the membrane
Rc = hydraulic resistance of the deposited cake
Since R, is a constant which can be calculated from the pure water flux, R,
can be calculated from the experimental flux. Figure 3.30 is a plot of the cake
resistance (R,) as a function of stirrer speed (in a stirred cell), protein concentra-
tion, and pressure. As the stirrer speed increases, the boundary layer thickness
decreases (see Figure 3.25), thereby increasing theconcentration gradient (dc/dx)
for removal of the cake. Lower protein concentrations in the bulk (Cb) also in-
crease the concentration gradient; the gel concentration at the membrane sur-
face is fixed (C,).
Effect of Pressure. One of the curious aspects of data like those in Figure
3.30 is that flux does not increase monotonically with pressure. Indeed, when
flux is plotted versus pressure, as in Figure 3.31, the flux often becomes inde-
pendent of pressure in the steady state.
When the pressure is increased, the flux does increase initially. The increase
results in a higher rate of convective transport of solute to surface of the mem-
brane. If the system is not “gel-polarized”, the solute concentration at the sur-
face (C,) increases resulting in an increase in the concentration driven back dif-
fusive transport away from the membrane. In fact, Cs will increase until the back-
diffusive transport of solute just equals the forward convective transport.
3000 \
UM-I MEMBRANE RESISTANCE, R,=350 IN.Hq-MlN/CM
PROTEIN
64
32
24
32
I2 0.20
32 0.75
26 0.60
\
\
\
\
’ -0.65 12 0.32
(RR.)STIR.
4000 l0,000 25,000 so,ooo l30,000
I I
lOI I I 8
300 1000 5000 1qooo
STIRRER SPEED,S (RPM)
Figure 3.30: Gel (cake) resistance as a function of stirrer speed, protein

centration and pressure.
0.10* I 1 I I I I
0.65% PROTEIN (1630 RPM)
TRANSMEMERANE PRESSURE (psi)

Figure 3.31: Effect of pressure on flux (flux becomes independent Of Pressure).
Ultrafiltration 169
Eventually the concentration at the membrane surface will be high enough

for a gel to form (C, = C,). Further increases in pressure will again temporarily
increase the convective transport (J,) to the membrane surface. However, since
the surface concentration is at a maximum, the back diffusive transport will be
fixed (assuming no changes in the fluid dynamics in the boundary layer), and so-
lute will accumulate on the membrane. The gel layer will thicken or compact
just compensating for the increased driving force (API by an equal increase in
the resistance of the cake (R,). The net result is that the flux will decrease to its
original value in the steady state. Therefore, in the “gel-polarized regime”, flux
is independent of pressure and is solely determined by the back-diffusive trans-
port.
In the steady state, the convective transport to the membrane must equal
the back-diffusive transport away from the membrane.
dc
JC= -Ds
where J = solvent flux through the membrane

C = concentration of membrane retained solutes or colloids
D = solute diffusivity
X = distance from the membrane surface
In the gel-polarized regime, the boundary conditions are fixed:
C = Cb at large distances away from the membrane

C = C, at the membrane surface.
Integrating Equation 9 assuming that the diffusivity (D) is constant
(IO) J
where K is the mass transfer coefficient:
D
(11) K-
?7
and 6 is the boundary layer thickness.

It is recognized that the diffusivity (D) is really a function of the concentra-
tion profile near the membrane surface. If one accepts an exponential mode of
the diffusivity:
(12) D = D,exp(-a Cl
where D, = diffusivity at infinite dilution

(II = constant
If one integrates Equation 9 taking into account Equation 12, the final
equation is considerably more cumbersome than Equation 10 without much gain
in accuracy.23 Suffice it to say that the calculation of flux from Equation 10 will
overestimate the flux slightly because the diffusivity in the boundary layer is
lower than that at infinite dilution (De).
Equation 10 shows that the flux is independent of pressure in the gel pol-
arized regime. Figure 3.31 shows that below some “threshold pressure” (Pr), flux
still increases with pressure. Lower solute concentrations (Cb) have higherthresh-
old pressures (Pt). Low solute concentrations favor a high back-diffusive trans-
port of solute, and a much higher flux is required to transport enough solute to
the membrane (JC) to begin to form a gel. This also means that the asymptotic
flux in the gel-polarized region of Figure 3.31 is higher for lower protein concen-
trations; the concentration gradient (C,-Cb) driving back-diffusive transport is
higher.
Likewise, at the same concentration, higher stirrer speeds result in a higher
asymptotic flux, because the boundary layer thickness (6) has been reduced-
increasing the mass transport coefficient (K) (see Equations 10 and 11). This
also means that the threshold pressure (Pt) will be higher for higher stirrer speeds.
The higher removal rate of solute from the membrane requires higher pressures
and flux to carry enough solute to the membrane to form a gel.
Once the gel-layer is formed, it is often the limiting resistance to flow. Fig-
ure 3.32 shows two membranes with widely different membrane resistances (I$,.,).
The pure water flux differs by a factor of 3.75; yet in the presence of protein
(retained by both), the water flux differs by a factor of only 1 .I 1, for pressures
over the threshold pressure of 20 psi. It will be noted that the higher hydraulic
permeability of PM 30 membrane results in a much lower threshold pressure (7
psi).
0.030 I , I 5 8 I 5 IO.5
A *
- 8.73
- 7.0
lRANSM&BRANE PRESSURE (psi)
Figure 3.32: Limiting resistance of gel layer (VS membrane resistance).
Of course, Equation 10 applies to all forms of fluid management-cross-flow

as well as stirred cells. For large scale systems, cross-flow techniques are pre-
ferred-where the feed stream flows tangential to the membrane surface (see
Ultrafiltration 171
Figure 3.33). There is a pressure drop down the channel or tube which means
that the first part of the channel may be gel-polarized; while the exit region may
not be. (Usually, a restrictor is placed on the exit retentate stream to keep the
exit pressure high so as to maximize flux throughout the channel length.) For
laminar flow, the entrance region of the channel may not be gel-polarized ei-
ther, because the boundary layer is not well developed at this point.
ULTRAFILTRATE
MEMBRANE
BOUNDARY
LAYER
FEED IN _+ RETENTATE
OUT
BOUNDARY
LAYER
ULTRAFILTRATE /
Figure 3.33: Cross-flow UF system.
Effect of Concentration. If Equation 10 applies, it should be possible to

plot the solvent flux versus the logarithm of the concentration (Cb) and get a
straight line with a negative slope equal to the mass transfer coefficient (K):
(13) J = Kin (C,) - Kin (C,j
The form of Equation 13 has been demonstrated for a large number of

macromolecular solutions and colloidal suspensions. The data of Figure 3.34
show the semilogarithmic variation of flux with concentration for two proteins
and two colloidal suspensions. According to Equation 13, the intercept with the
abscissa should occur when Cb = C,. This provides a way of experimentally de-
termining the gel concentration. The values obtained from Figure 3.34 are en-
tirely reasonable. Many protein solutions reach their solubility limit between 25
and 35%. Likewise, for colloidal suspensions, the equivalent of a gel layer should
be a layer of close packed spheres having a packing density between 60 and 75%.
The negative slope of each of the straight lines in Figure 3.34 is equal to the
mass transfer coefficient which is a strong function of the stirrer speed or the
tangential velocity across the membrane. Figure 3.35 shows increasing slopes
with increasing recirculation rate (the volumetric flow rate of retentate recir-
culated back as feed to the channel, which is proportional to the tangential ve-
locity across the membrane). All lines converge at zero flux where the concentra-
tion equals the gel concentration (C, = 28%). Experimentally, it is difficult to
carry out a concentration at constant cross-flow velocity (recirculation rate); the
viscosity increases with concentration requiring constant adjustment of the
pump. Some data in the literature show flux vs. log concentration curves. Often
the reason for the curvature is the inconstancy of the recirculation rate; as the
280 I I IIll\l I I III1
260 Code
o Electra deposition primer
-=
i 240 A Styrene butodiene latex
5 220 o Human albumin
5 200 l Gelatin (@I 7O’C)
180
160
140
120
100
80
60
40
20
0
1 2 3 4 5678910 20 30 40 50 80
Concentration, wt % solids
Figure 3.34: Semilogarithmic variation of flux with solute concentration.
1 I I I I1111 I I III -
140 Ultrof iitrotion of human albumin
II .
1 I I I I III11
1 2 3 4 5 678910 20 30 40: 50 60
Protein concentration (wt. %I
Figure 3.36: Effect of recirculation rate (cross-flow velocity) on flux variation

with concentration.
Ultrafiltration 173
process stream is concentrated, the recirculation rate decreases, and the flux
drops to lower operating lines (see Figure 3.35). When the data are extrapolated
to zero flux, the inconstancy of recirculation rate can lead to inaccurate esti-
mates of the gel concentration (Cc).
In addition, if the data are not gel-polarized, Equation 13 will not plot as a
straight line on semilog paper. The concentration at the surface of the mem-
brane (C,) is less than C, and not constant. Therefore, at low pressures or low
concentrations, some curvature is expected. In Figure 3.36, protein concentra-
tions below 1% at high recirculation rates are not fully gel-polarized; the back-
diffusive mass transport removes protein from the membrane surface at a high
rate. Low recirculation rates K2.5 GPM) have lower back-diffusive transport
and are gel-polarized. All data on Figure 3.36 could be gel-polarized if the av-
erage transmembrane pressure were raised sufficiently. However, at extremely
low solute concentrations, there is a ceiling on solvent flux due to the limiting
hydraulic resistance of the membrane; eventually, the flux will reach an asymp-
totic value equal to the pure solvent flux of the membrane.
l-
, Recirculation rate
I 11111 I I I111111
1.0 2 4 6 8 10.0 20 40 60 80 100
Protein concentration (wt %)
Figure 3.36: Variation from semi-log dependence in non gel-polarized regime.
Effect of pH. Comparing Figures 3.35 and 3.36, both showing flux data on
human albumin solutions, it will be seen that the gel concentration is 28% in one
case and 45% in the other. This reflects the difference in solubility of proteins
for different conditions such as pH.
Since the isoelectric point of most proteins is between a pH of 4 and 5, it
would be expected that operation in this pH range would result in the lowest
gel concentration. Further, since changes in pH are not expected to change the
mass transfer coefficient (K), the operating lines in Figures 3.35 and 3.36 do
not change in slope but are simply shifted to the left or right depending on the
gel concentration (Cc). Thus, for a given concentration, there can be dramatic
change in flux with pH.
Figure 3.37 shows the variation in the UF flux from cheese whey with pH.
Notice the minimum at the isoelectric point.
l3-
12 -
I1 -
g lo-
5
EL
- 9-
ii
c
9-
t
7 - Minimum associated
with
isoelectric paint
62 I
3 41 5 6 71 9,
‘PH
Figure 3.37: Effect of pH on flux (Cheshire cheese whey).
Evaluation of the Mass-Transfer Coefficient

It is clear from Equation 10 that the UF flux is determined largely by the
mass-transfer coefficient (K). With proteins and other polyelectrolytes, we can
modify the gel concentration (Cc) by altering the pH and/or ionic strength of
the medium. However, for most process streams, that option is not permissable.
Therefore, the optimization of flux is largely effected through the parameters
that effect the mass transfer coefficient.
The mass transfer-heat transfer analogies well known in the chemical en-
gineering literature make possible an evaluation of the mass transfer coefficient
(K) and provide insight into how membrane geometry and fluid-flow conditions
can be specified to optimize flux.%
Laminar Flow. The Graetz or L&que solutions25r26 for convective heat
transfer in laminar flow channels, suitably modified for mass transfer, may be
used to evaluate the mass transfer coefficient where the laminar parabolic veloc-
ity profile is assumed to be established at the channel entrance but where the
concentration profile is under development down the full length of the channel.
For all thin-channel lengths of practical interest, this solution is valid. Levgque’s
solution26 gives:
Ultrafiltration 175
o‘aa
(14)
for 100 < Re SC : < 5000
where Sh = Sherwood Number = Kdh/D

Re = Reynold’s Number = Udh/v
SC = Schmidt Number = v/D
dh = equivalent hydraulic diameter
L = channel length
u = average velocity of fluid
V = kinematic viscosity = p/p
cc = viscosity of fluid
P = density of fluid
or
(15)
for flat rectangular channels, where dh = 2b; (b = channel height).
(16)
or
(17)
where Q = the volumetric flow rate

W = the channel width
More generally,
. o.aa
(18) K = 0.816 ; 02
( >
where i = the fluid shear rate at the membrane surface

= 6U/b for rectangular slits
= 8Uld for circular tubes
A review of Equations 14-18 shows that the flux (or mass transfer coeffi-
cient) may be increased by increasing the channel velocity (U or Q) or by de-

creasing the channel height (b). In more general terms, any fluid management
technique which increases the fluid shear rate (i) at the membrane surface will
increase the flux.
Indeed, Equation 18 shows that in laminar flow, for a fixed bulk stream
concentration (Cu), the flux should vary directly as the cube root of the wall
shear rate per unit channel length. This has been confirmed for a large number of
solutions ultrafiltered in a variety of channel geometries (see Figure 3.38).
0.3 I , / . - * * . *. 140
_ 0.2.
c
E
I 0.1 -
“6 0.08 -
& 0.06 -
-7 =
s
3
- 2:
- 2.1
- 1.4
- 0.7
I I
100 1000
WALL SHEAR RATE
6RARREL LENGTH 1 f/L9 (set-cm)-’
Figure 3.38: Experimental confirmation of 0.33 power dependence of flux on

wall shear rate/channel length (in laminar flow).
In laminar flow, the channel length (L) is important because the boundary
layer develops as fluid moves down the channel. Longer channels have a thicker
boundary layer (at the end) which reduces the back diffusive transport. Conse-
quently, the average flux is lower. This would argue for shorter channel lengths.
In practice, this is not cost effective since a significant portion of the cost of
manufacturing a module is associated with potting or connecting the chan-
nels/tubes to a common “header”. Short channel lengths mean a high cost per
unit of membrane area.
Turbulent Flow. Perhaps the best known heat-transfer correlation for fully
developed turbulent flow is that owing to Dittus and 8oelter.27 The mass trans-
fer analogy based on the Dittus-Boelter correlation is:
(19) Sh = 0.23 Re0.8 SCO.~~

Ultrafiltration 177
(20) K = 0.023
It can be argued that any turbulent flow correlation should not be applied
for Re <lO,OOO. However, in current thin-channel ultrafiltration devices, the en-
trance geometry is such that fully developed turbulent flow occursat much lower
Reynold’s numbers. Measurements of fluid velocity versus pressure drop show a
definite transition from laminar to fully developed turbulent flow at Re = 2000.
For flat rectangular channels where dh = 2b, Equation 20 becomes:
(21)
or
(22) K = 0.02
As in laminar flow, the flux (or mass transfer coefficient) may be increased
by increasing the channel flow rate (Q) and by decreasing the channel height (b).
The effect is, however, much more dramatic in turbulent flow-the flux varying
with flow rate to the 0.8 power and inversely with channel height. Furthermore,
the flux is independent of channel length since both the velocity and concentra-
tion profiles are established rapidly in the entrance region of the channel. Again
as in laminar flow, most of the ultrafiltration data taken on solutions in turbu-
lent flow are in good agreement with theory, i.e., with the Dittus-Beelter corre-
lation. For example, the turbulent flow data of Figure 3.39 show a 0.75 power
dependence of flux on recirculation rate (Q) in good agreement with the pre-
dicted 0.8 power. (The IO-mil channel data were taken in laminar flow, whereas
the 30-mil channel data were taken in turbulent flow.)
The laminar flow data of Figure 3.39 have a higher slope (0.52 than pre-
dicted by theory (0.33)-probably because of “secondary flow effects”. The
data were taken in a spiral flow thin channel device. Whenever fluid passes
through a curved tube or channel, centrifugal forces tend to throw fluid outward
from the center of the channel. It then recirculates inward along the walls of the
channel (see Figure 3.40). It is well known that coiled tube heat exchangers pos-
sessessuperior heat transfer characteristics because of secondary flow effects.
At small curvatures, the Dean number (De) governs the transport processes
in coiled tubes and channels:
De = Re J +
where a and R are tube and coil radii respectively. Dravid et al’e has shown ex-
perimentally that the heat-transfer coefficient in a coiled tube varies as De0’55 in
the fully developed region of the boundary layer. By analogy, the mass-transfer
coefficient for laminar flow in a spiral channel should vary as Re”” rather than
as ReoeJ3.This is precisely what occurs in the data of Figure 3.39.
1 , I 1 - .I.‘,
10 NIL CHANNEL
ULTNAFlLTRATION
OFHUHAN
ALWIIN
T&C - PN30NfNBNANE
NECIRClNATION
NATEkC/HIN)
Figure 3.39: Experimental confirmation of 0.8 power dependence of flux on

recirculation rate (in turbulent flow).
Outer
wall
Figure 3.40: Secondary flow patterns in helically coiled tube.

Ultrafiltration 179
Theoretical Prediction of Flux

The success of the L$que and Dittus-Boelter relationships in indicating
the variation (power dependence) of ultrafiltrate flux with channel geometry
and fluid velocity for macromolecular solutions is gratifying. The more Crucial
test of the theory, of considerable interest to the design engineer, is whether
these relationships can be used to calculate quantitatively the Ultrafiltrate flux
knowing the channel geometry, fluid velocities and solute characteristics.
Macromolecular Solutions. Figure 3.39 shows that agreement between the-
oretical and experimental values is good in at least some cases (i.e., spiral flow
channel plates). Equation 17 for laminar flow and Equation 22 for turbulent
flow were used to calculate the mass-transfer coefficient (K). The diffusivity of
albumin (D = 6 x IO7 cm’/sec) was obtained from a handbook.29 The gel con-
centration (C,) was determined experimentally from data like those in Figure
3.35. Although the slope in laminar flow was 0.52 compared with the theoreti-
cal slope of 0.33, the experimental values were still well within 25% of the theo-
retical values. Agreement in the turbulent regime is more striking. The accuracy
of the L&&que and Dittus-Boelter relationships has also been verified in linear
thin-channel tubular equipment ultrafiltering protein solutions. Figure 3.41 pre-
sents laminar flow data from 15mil channel tubes compared with theoretical
values using a diffusivity of 6 x lo-‘cm’/sec and a gel concentration of 45% as
determined from Figure 3.36. Again, experimental and theoretical values agree
within 25% although the theoretical values are consistently lower.
ULTRAFILTRATION
OFWANALWilN
LTC-1WI%)(15 MIL CHANNEL)
30 P-SIG AVG. TRANSMEWBR!NE PGESSUBE DROP
GECIRCUIATION GATE (GPtl)
l.OL I L . . ..a.
1 2 3 4 5 6 7810 20 30 4050
Figure 3.41: Theoretical vs experimental values of flux with recirculation rate

(in laminar flow).
The ability of the L&&que solution to adequately describe the variation in

flux with diffusivity of the retained solute is illustrated in Figure 3.42. Here,
albumin data were compared with whole serum data. The larger. globulins in
whole serum have a lower diffusivity D = 4 x IO1 cm2/secz9 and a lower solubil-
ity limit (C,). The theoretical curves are 15 to 20% below the experimental data
in both cases. Thus, the dependence of flux on diffusivity to the 0.87 power is
confirmed.
5-u CONCENTRATION OF SERUMPROTEINSIN

LTClS(15HlL CHANNEL)(PbfiO MHBRANE)
RECIRCULATION RATE 23 GPH
P,, - 65 PSI Poll, = 20 PSI
T- 23'C 1
40
30
20
10
PROTEINCONCENTRATION
WGT.X
I I I III. I
01.7 1.0 2 3 45 6 7 8 910 20 30 40 !a
Figure 3.42: Variation in flux with diffusivity of retained solute.
Other macromolecules in solution seem to fit the gel-model as well as pro-

tein solutions. For example, data from the literature3’ on the ultrafiltration of
polyethylene glycol (Carbowax 20 MI solutions in one inch diameter tubular
membranes operating in turbulent flow are plotted along with theoretical flux
values, in Figure 3.43. The theoretical values were calculated using a diffusivity
of 5 x lOI cm2/sec3’ and a gel concentration of 7.5%.30 The agreement between
theoretical and experimental values is within 14 to 27% at the higher and lower
Reynold’s numbers, respectively.
Colloidal Suspensions. The agreement between theoretical and experimental
ultrafiltration rates for macromolecular solutions can be said to be within 15 to
30%. For colloidal suspensions, experimental flux values are often one to two or-
ders of magnitude higher than those indicated by the L&e^que and Dittus-Boelter
relationships.
Ultrafiltration 181
HFA-200 Tubular Membrane (I” Diameter)

0.9 W Carbowax 20 M Feed
Figure 3.43: Theoretical vs experimental values of flux for 0.9% polyethylene

glycol solution.
Equations 17 and 22, for laminar and turbulent flow respectively, both indi-
cate that the mass transfer coefficient should vary with the diffusivity of the re-
tained solute to the 0.67 power. That this is the case for macromolecules in solu-
tion was shown in Figure 3.42.
The Stokes-Einstein relation for diffusivity is:
(24)
where k = Boltzman constant (i.e., the molar gas constant divided by

the Avogadro number) = 1.380 x IO-l6 erg deg’
T = Absolute temperature “K
/J = viscosity in poise
ro = radius of the particle diffusing in cm
Macromolecules of higher molecular weight will have larger particle dimen-

sions and lower diffusivities. For example, protein diffusivities of interest in this
chapter are as follows:
Protein M.W. D 20°C cm2/sec.
Albumin 65,000 6 x 10-7
y globulin 170,000 4 x 10-7
Collagen (gelatin) 345,000 0.7 x 10 -7
These proteins should exhibit decreasing flux rates in the order given, and
this can be seen in Figures 3.34 and 3.42. Referring to Equation 24, one would
expect that the ultrafiltration rate from whole blood, including the 8 /.r red cells,
should be considerably less than that from the plasma above. That such is not
the case is illustrated in Figure 3.44.
al _ II I I III I I II,,,,,
- WHOLE BLOOD AND PLASMA
z UM-IO MEMBRANE, AP= 10 psi
ESMOND CELL, L=l3 cm
9 ___-------------- WATER FLUX
“E
-5
0 WHOLE PLASMA-35 mils

0 WHOLE BLOOD-35 mils
o WHOLE PLASMA-17 mils
n WHOLE BLOOD- I7 mils
A WHOLE PLASMA-S mils
I I I.1111 I I Illll I I III

2 4 6 a 2 4 6 a 2 4 6 6
IO’ I02 IO3 I04
AVERAGE SHEAR RATE AT MEMBRANE SURFACE (set”)
Figure 3.44: UF of whole blood and plasma.
Likewise, the polymer latex data of Figure 3.34 have a much higher slope
(mass transfer coefficient) than would be expected from the latex particle dif-
fusivity. Monodisperse polystyrene latexes have both suspension viscosities and
Ultrafiltration 183
sedimentation coefficients in good agreement with the predicted Stokes-Einstein

diffusivity. The Stokes-Einstein diffusivity was calculated to be 2.3 x 10-s cm*/sec
for a carboxylic modified styrene-butadiene copolymer latex which had an aver-
age particle size of 0.19 p. This is in good agreement with other values reported
in the literature.
The gel concentration of this material has been determined to be 75% for
numerous thin-channel runs in both tubular (see Figure 3.45) and spiral flow
equipment.
Using the values cited above for the diffusivity (D) and the gel concentra-
tion (C,), the theoretical flux is plotted for the 40 GPM recirculation rate in
Figure 3.45 (thin-channel tubes) and found to be l/38 of the experimental thin-
channel tube value! Even if the calculated diffusivity were an order of magnitude
larger, the theoretical flux would still be less than 5/sof the experimental value!
13c-
ULTRAFILTRA.TION
OF STYRENEBUTADIENE
no- \ POLYMERLATEX IN LTC-I
\ 15 NIL CHANNELS- XEM WIBRANE
RECIRCUldTlON
RATE 40 GPN
llflr
-_ '\ GO PSIG MR&%_TIW'E"G~NE
\
\
100- \
\
9Q-
80-
70 -
GO-
50-
40-
30 -
20 -
rn-
1”
CONCENTRATION
WOL XI THEORETICAL 40 GPH
1 2 3 4 5 678910 20 30 40 50
Figure 3.45: UF of styrene butadiene polymer latex showing disparity between

experimental results and theoretical prediction.
Flux vs. recirculation rate data for a thin-channel tubular unit are plotted in
Figure 3.46. Theoretical values are also plotted assuming laminar and turbulent
flow. Most of the data on this plot (below 30 GPM) are taken at Reynold’s
numbers below 2000 and therefore assumed to be in laminar flow. The discrep-
ancy between the experimental flux value and theoretical laminar flow value is a
factor of 38 at 40 GPM and a factor of 15 at 5 GPM! If the assumption is made
that these data were taken in turbulent flow (completely unwarranted), the
slope is more nearly that indicated by theory, but the discrepancy is still a factor
of 7.5 at all recirculation rates.
ULTNAFILTNATIONOF STYRENEBUTADIENEPOLYMER
101l- LATEX IN LTC-1 (15 NIL CHANNEL)
XMSO
81
61
41
21
11
I
/- RECIRCULATION
RATE (GPM)
1 I
1 2 3 4 5 6 78910 20 30 40 70
Figure 3.46: Flux vs recirculation rate in linear thin channels (styrene butadiene
polymer latex).
The gross discrepancy between theoretical and experimental flux values also
exists in data obtained from spiral flow thin-channel equipment (see Figure
3.47). In these data, the transition from laminar flow is clearly seen (change in
slope) at Reynold’s numbers only slightly below 2000. However, there is no
abrupt change in flux, as calculated from theory, suggesting that the data labeled
“turbulent flow” may be in a more nebulous transition region between laminar
and turbulent flow. The experimental slopes tend to be higher than the theoreti-
cal slopes in both laminar and turbulent flow except for the laminar flow data at
1% concentration which follows the predicted ‘h power dependence. Again, the
gross failure of the theory in estimating experimental values is evident. The ex-
perimental laminar flow values are a factor of 19-29 higher than the theoretical
values whereas the experimental turbulent flow values are a factor of 8-15 higher
than the theoretical values.
Ultrafiltration 185
I I Illll I I
I
ZOC- ULTRAFILTRATION OF STYRENE BUTADIENE

LATEX TCl-D (30 NIL CHANNEL)
(PM30 WIBRANE) 40 PSI AVG.
TRANSMEHBRANE PRESSURE DROP
100 200 400 600 1000 2000 4000
Figure 3.47: Flux vs recirculation rate in spiral thin channels (styrene butadiene
polymer latex).
An evaluation of more than 40 different colloidal suspensions in our labora-

tories has indicated that the diffusion coefficient calculated from the ultrafil-
trate flux using the L&v$qqueor Dittus-Boelter relationships is generally from one
to three orders of magnitude higher than the theoretical Stokes-Einstein diffusiv-
ity.
It is evident from the above that minor adjustments in molecular parameters
such as diffusivity, kinematic viscosity, or gel COnCentratiOn are incapable Of ra-

solving order of magnitude discrepancies.
Similar discrepancies were noted by Blatt et a13’ for colloidal suspensions
such as skimmed milk, casein, polymer latexes, and clay suspensions. Actual ultra-
filtration fluxes are far higher than would be predicted by the mass transfer co-
efficients estimated by conventional equations, with the assumption that the
proper diffusion coefficients are the Stokes-Einstein diffusivities for the primary
particles. Blatt concluded that either (a) the “back diffusion flux” is substan-
tially augmented over that expected to occur by Brownian motion or (b) the
transmembrane flux is not limited by the hydraulic resistance of the polarized
layer. He favored the latter possibility, arguing that closely packed cakes of col-
loidal particles have quite high permeabilities. However, this is not a plausible
hypothesis for the following reasons:
(1) If the gel layer is not the limiting resistance to flow, the membrane
must be, and the flux should be proportional to the transmem-
brane pressure drop. Experimental data deny this-showing thresh-
old pressure (above which flux is independent of pressure) (see Fig
ure 3.48).
(2) If the gel layer is not the limiting resistance to flow, the layer will
continue to grow until the channel is completely full of 75% solid
material-resulting in a drop in recirculation rate with time. Thin
channel tubes and spiral flow modules running continuously at
constant latex feed concentration and pressure drop for periods ap-
proaching one year have shown no decreases in recirculation rate or
accumulation of polymer latex in the channels.
These observations lead to the conclusion that the back-diffusive transport

of colloidal particles away from the membrane surface into the bulk stream is
substantially augmented over that predicted by the L&v$que or Dittus-Boelter
relationships. It is known that colloidal particles flowing down a tube tend to
migrate across the velocity gradient toward the region of maximum velocity; this
is called the “tubular pinch effect”.
Tubular Pinch Effect. The lateral movement of particles across the stream-
lines in laminar flow was first observed and recorded in 1836 by Poiseville. He
noted that the region immediately adjacent to the walls of blood capillariestends
to be free of blood cells.
This phenomenon also explains why many colloidal suspensions exhibit less
frictional pressure drop than would be expected from the fluid viscosity. The
apparent viscosities of such suspensions vary with tube radius, length, and flow
rate.33 To account for such anomalies it has been postulated that there exists a
lubricating particle-depleted (“plasmatic”) layer at the wall of vessels in which
there is a nonuniform shear field. For example, blood flowing through fine glass
capillaries reaches an equilibrium state in which the red cell concentration in the
tube is less than that in the inflowing or outflowing blood-presumably the result
of axial drift of red cells and their consequent faster average transit than plasma.
Palme? was able to skim off a plasma rich layer at thewall through fine branches
and was able to measure increases in hematocrit from near the wall to the axial
region.
Ultrafiltration 187
11
I
VARIATIONOF FLUX WITH TRANSMMMNE PRESSUREDROP
STYRENEBIJTADIENE
POLYt'!ER
LATEX IN LTC-1
90 (1s UlL CHANNEL~MiM IIEMBRANEI
ATREIATIVELYLOW RECIRCULATION
RATE - 11.8 GPH
1X LATEXCONCENTRATION
AVERAGETRANSREH8RANE
PRESSUREDROP (PSI)
10 20 30 40 9 GO 70 80 90
Figure 3.48: Independence of flux on pressure (styrene butadiene polymer latex).
Segr; and Silberberg35~36-working with dilute suspensions of rigid spheres,

were the first to publish their observations of the “tubular pinch effect”, where
the particles migrated away both from the tube wall and the tube axis reaching
equilibrium at an eccentric radial position. At this position, the spheres became
regularly spaced in chains extending parallel to the tube axis. The observations
of Segr6 and Silberberg have spawned a number of theoretical and experimental
studies investigating the effect and analyzing the cause. The “tubular pinch ef-
fect” has also been observed experimentally for suspensions of rubber disks,
carbon black, polystyrene spheres, PVA spheres, aluminum-coated nylon rods,
elastomer filaments, aluminum particles, insoluble salts, glycerol and silicone oil
in various continuous flowing media.
One of the more spectacular visual studies was made by Brandt and
Bugiarello.37 They made direct photographic observations of small Dylite spher-
ical beads suspended at concentrations of 1.7 to 5% in a glycerin-water solution
and flowing in laminar flow (Re from 400 to 1640) through long rectangular
channels made of Plexiglass. The ratio of the channel width to the bead diameter
was 25.6. Figure 3.49 gives the average half-channel particle distribution (ob-
tained photographically) expressed as the ratio of the number of particles per
band (C,) to the total number of particles (C total) in the entire channel width.
.
Flow of Monolayer-s of Suspended Particles
Ul..UL1
‘yp
-___-_ - c, WIFOI” l
,c 101.L
‘! IY.50”1*
- . .+.I
-_____-___-_
‘Y’.
L /;
__________
-__..__--.
I- 0
IO
.
_.
I
m
J
I
L’: =I
0 L-.-* ..-
Ir.
s-
I
c.
__________ - ----_------
!
C
O- .- I
1
P
Ly-207
5- i
----w-w ____________
’ 4
____
IO L
I
0’ I
5.
O-
S
L&.;_,I
-----
---
EEJ____
E_:_:.
L4 e--2._._
; l
____ __
!
0 i
I
0.20 0.30 0.40 0.54
x
hrhdc disl.ributions. c.,.# = &(.I%; Q = 20 cm3 ~CC-1.
Figure 3.49: Particle migration to center line of flowing channel (tubular pinch
effect). Brandt and Bugliarello (1966).
Ultrafiltration 189
The plots show very clearly the migration of particles to the center of the chan-
nel as the distance (Y) from the inlet of the channel increases. Brandt and
Bugliarello found that the process was accelerated by increases in flow rate and
delayed by increases in average concentration.
Colloidal particles are large enough that any given particle, not in the cen-
ter of the tube, will experience a pressure difference across its diameter due to
the higher velocity on one side than on the other. Thus, the particle will exper-
ience a “lift-force” tending to move it away from the wall toward the center of
the channel. This “lift-force” is not unlike the aerodynamic lift on an airfoil due
to the “Bernoulli effect”.
The Navier-Stokes equation has been solved By Cox and Brenner (unpub-
lished data) who computed the lateral force required to maintain a sphere at a
fixed radial position (t-1.They used Stoke’s law for the neutrally buoyant case:
(25)
where V is the radial migration velocity

U is the average fluid velocity down the channel or tube
Re is the Reynold’s number
t-o is the particle radius
R is the tube radius
r is the radial position of the particle in the tube
F(r/R) is a function of the radial position of the particle in the tube
or channel.
Equation 25 has the same form as the empirical equation used by Segre and
Silberberg% to correlate their data:
where r* is the equilibrium radial position of the particle which was found
to decrease as (m/R) increased.
The tubular pinch effect can explain much of the anamolous UF data for
colloidal suspensions. With UF, the water flux through the porous wall will still
carry particles to the wall, but the “lift” of particles away from the wall (due to
the tubular pinch effect) will certainly augment the back diffusive mass transfer
described by the L&e^que and Dittus-Boelter relationships.
Equations 25 and 26 both predict a radial migration velocity (VI increasing
as the square of the cross-flow velocity (U) [The Reynolds Number (R,) also in-
cludes U). This can explain why the dependence of UF flux on cross-flow veloc-
ity (U) can be higher than 0.33 in laminar flow and 0.75 in turbulent flow (e.g.,
see Figures 3.50 and 3.5 1).
LAMINAR FLOW
79 ELECTRODEPOSITION
PRIMER
VARIATION OF FLUX WITH
RECIRCULATION RATE
mGw
25
Figure 3.50: UF of electrodeposition primer (0.85 power dependence on recir.

culation rate in laminar flow).
3-
j- TURBULENT FLOW
15% ELECTRODEPOSITlUN
PRIMER
VAAIATION OF FLUX WIT" /
RECIRCULATION BATE 0
UGMJ al
0
00 /
SLOPE=l.234
TC-1 (30 mil Channel)

/ PM30 Membrane
0
0’ 0
I * . . I I * , , I I , *, (
1 2000 2500 3000 3

RECIRCULATION RATE (mL/MIN)
Figure 3.51: UF of electrodeposition primer (1.23 power dependence on recir-

culation rate in turbulent flow).
Ultrafiltration 191
The data of Figure 3.44 show similar flux values for whole blood and plasma.
Presumably, the tubular pinch effect tends to depolarize the membrane surface
of red cells yielding a flux similar to that obtained with plasma alone. In some
cases, the flux with the red cells present is higher than that with plasma alone.
The migration of the larger red cells away from the membrane surface tends to
drag the plasma proteins along.
Bixler et a13* found that adding glass and plastic beads (ranging in size from
30 to 100 /J) to a protein solution augmented the UF flux (see Figure 3.52). He
attributed this to (1) the mixing action of the particles and (2) the mechanical
scouring of the membrane surface.
I+ I I I I I I
6-
8..
0 5 10 15 20 25 30 40
PARTICLELOADINGCVOL,%)
Figure 3.52: Enhancement of protein flux with particle addition.
Equations 25 and 26 also predict that the radial migration velocity (V) will
increase as the tube radius (R) decreases. Thin channels are more effective in de-
polarizing the membrane surface via the tubular pinch effect. This may explain
the larger discrepancies between experimental and theoretical flux values in 15
mil channels (see Figure 3.46) than in 30 mil channels (see Figure 3.47).
Green and Belfort3’ have combined the equations for particle migration due
to the tubular pinch effect with the normal back-diffusive transport to calculate
flux values which are closer to the experimental results. A step-wise iterative
procedure was used to calculate a detailed particle trajectory analysis. Unfor-
tunately, these complex theoretical calculations still come short of accurately
predicting the experimental flux from various colloidal suspensions.
Augmented Cross-Flow Effects

Other methods have been sought to augment the depolarization of the mem-
brane by cross-flow. For particles with a higher density than water, centrifugal
force has been explored. For particles bearing an electric charge, an electric field
has been investigated.
Centrifugal Force. Robertson et aVk’ designed an apparatus in which the
centrifugal force vector was perpendicular to the membrane surface but opposite
(and parallel) to the flux vector. Solutions of casein and dextran (60,400 dal-
tons) were passed over the membrane in laminar flow. When the apparatus was
spun, centrifugal field strengths from 100 to 600~ resulted in flux improvement
factors of 3 to 16.
Figure 3.53 shows the increase in the mass transfer coefficient (K), as cal-
culated from Equation 10, due to the centrifugal force. The family of curves are
for different concentrations of casein.
8.64%
I .81X
I .6BX
3.eex
7.85%
0 Ia0 200 380 400 580 600
Centrifugal Force (x g)
Figure 3.53: increase in mass transfer coefficient (K) with the imposition of
centrifugal force.
Electric Field. Many particles and/or macromolecules bear a charge (USU-

ally negative). Consequently, the application of an electric field can cause these
species to migrate away from the membrane, thereby augmenting the mass trans-
fer coefficient (K). Henry et a141 has investigated cross-flow electro-filtration of
kaolin clay suspensions and oil-water emulsions. Since both are negatively charged
in aqueous suspensions, an electric field will always give higher filtration rates
than cross-flow filtration alone (see Figures 3.54 and 3.55). The increase in UF
flux depends on the electric-field strength as well as the cross-flow velocity.
Ultrafiltration 193
OIL/WATER
2. -
300 PPM
20 #-
Figure 3.54: Cross-flow electro-filtration of oil/water emulsion.
cL*yhwar
500 PPM
op*1.05.dNlM2
--•-----h E: 394 VICM

E>E,
1 2
fq&d4 x103 5 6
Figure 3.55: Cross-flow electro-filtration of kaolin suspension,

One of the more curious phenomena associated with cross-flow electrofil-

tration is that above some critical voltage (E,), increases in the tangential veloc-
ity across the membrane may actually decrease the membrane flux as in Figures
3.54 and 3.55. This can be explained by referring to Figure 3.56.
Tangential Velocity
I
I ,
t I Electrophoretic
_
a) E < EC c Cake
I Bulk
)I
1 Electrophoretic
Particle
Migration
.A
Medium
-J
CB No Diffusive CS
Transport
b) E = E,
Electrophoretic -
Particle
Migration
Bulk
‘- J
CB
Diffusive Particle
Transport
c) E > EC
Figure 3.56: Three operating regimes for cross-flow electro filtration.

Ultrafiltration 195
There are three operating regimes in cross-flow electrofiltration depending

on whether the field strength (E) is above, below, or equal to the critical voltage
(E,). The critical voltage is defined as the voltage at which the net particle migra-
tion toward the membrane is zero, i.e., where the convective transport of parti-
cles toward the membrane is exactly equal to the electrophoretic migration of
particles away from the membrane.
Regime (a) in Figure 3.56 is for E less than E,. Concentration polarization
still exists, but the flux is higher than normal because the back-diffusive trans-
port of particles away from the membrane is augmented by the electrophoretic
particle migration. In this regime, increases in the tangential velocity will dimin-
ish the boundary layer thickness resulting in a higher flux.
At the critical voltage, regime (b), there is no concentration polarization be-
cause the electrophoretic transport is equal to the convective transport. In this
regime, increasing the tangential velocity is expected to have no influence on the
flux because fluid shear can only improve the transport of particles down a con-
centration gradient. In this case, there is no concentration gradient.
When the voltage is greater than the critical voltage, regime (c), the electro-
phoretic migration of particles away from the membrane is greater than the con-
vective transport to the membrane. Thus, the concentration of particles near the
membrane surface is depleted due to removal by the electric field. In this case,
increasing the tangential velocity decreases the flux, because the diminished
thickness of the boundary layer increases the diffusive transport of particles
down the conventration gradient toward the membrane (see Figures 3.54 and
3.55).
In Figure 3.55, the lower curve for E = 3.9 v/cm shows a transition in slope.
The flux decreases with decreasing Reynolds number until a point is reached
where the convective transport of particles toward the membrane is just equal to
the electrophoretic migration away from the membrane (i.e., the voltage becomes
critical as we decrease the cross-flow velocity). Further decreases in cross-flow
velocity will not decrease the flux as there is no concentration polarization.
Though cross-flow electrofiltration has not yet been commercialized, its
greatest potential appears to be in the elimination of the gel layer altogether, by
the application of voltages above E,. Solutions/suspensions which cannot be ul-
trafiltered economically due to severe fouling problems might be filtered with
this technique if the particles and/or macromolecules bear a charge. It is even
possible that this technique might make the fractionation of gamma globulin
from albumin feasible. Oil-water emulsions might be separated by higher flux
MF membranes. Without the electric field, the oil droplets would not be retained
by the larger pore sizes in MF membranes, but above the critical voltage, break-
through can be avoided. For example, the data of Figure 3.54 were taken using
a 0.6 /.r pore size membrane. (Normally, most oil droplets in emulsions are be-
tween 0.1 and 0.5 /.I in diameter.)
Effect of Temperature
It has been found experimentally for a large number of membrane systems
(including MF, UF and RO) and feed streams that the permeation rate is inversely
proportional to the fluid viscosity. Since the viscosity of water decreases by
about 2.5% for every “C rise in temperature, membrane researchers often refer to
the 3% rule (that flux increases 3% per “C) as a rough rule of thumb.
Pure Water Transport. These results are not unexpected for the transport of
pure water through porous membranes. Poiseuille’s law predicts that flow (J)
through any porous media should be described as follows:
Nnd4AP
(27)
‘= iz?jYL
where N = number of pores per square foot

d = pore diameter (average)
AP = pressure drop
p = fluid viscosity
!.? = pore length (including a tortuosity factor)
The only variable in Equation 27, which is temperature dependent, is the vis-
cosity:
where AE, is the activation energy (3750 Cal/g-mol for water).

However, as we have seen, the membrane itself is no longer the limiting re-
sistance to flow when “gel-polarized”. The flux is given by Equation 10 instead
of Equation 27. The mass-transport coefficient (K) and the gel concentration
(C,) vary with temperature.
Figure 3.57 is an illustration of how different protein solubilities at 4°C and
25°C will shift the gel-concentration. (Incidentally, the lower slope of the IOW-
temperature data was due to a lower recirculation rate due to the increased vis-
cosity of the process stream). Clearly, the inverse viscosity rule does not apply
for all points on the graph. At 45% solids, the ratio of the flux at 25°C to that at
4°C goes to infinity.
SOLIDS CONTENT (weight percent)

Figure 3.57: Effect of temperature on UF of a high-protein meal.
Ultrafiltration 197
On the other hand, some proteins seem to show a gel concentration invari-
ant with temperature (see Figure 3.58). In this case, the higher temperatures
tend to denature the proteins (reducing their solubility), thereby compensating
for the expected increase in solubility with temperature. Thus, the effect of tem-
perature on gel concentration is complicated and often unique for the solute/sol-
vent system being filtered.
9.r(cm2 of membrane, ll-mil

channels, recirculation rate
of 150 cdmin at 25 psig
IO -
50
CONCENTRATlOll?OF PROTEIN CW6?% 1
Figure 3.58: Effect of temperature on UF of bovine serum.
Differentiating Equations 16, 21, 24 and 28 yields the following expressions

for the dependence of flux on temperature:
(29)dT
For laminar flow gel polarization
dJW
For turbulent flow gel polarization
(30)
The derivation of Equations 28 and 29 may be found in Reference 17-Ap

pendix E, pp. 2-97 to 2-99. It assumes that C, is invariant with temperature.
When these equations are used to calculate the percent variation of flux with
temperature near room temperature, the results are usually within 25% of that
predicted by the inverse viscosity rule.
MEMBRANE FOULING; FLUX DECAY AND RESTORATION
With some process streams the flux can be stable for months or even years
without cleaning or membrane replacement. For most applications, however,
there is a gradual flux decay with time as in Figure 3.59. This is not due to inter-
nal pore fouling (as in symmetrical MF membranes). Rather it is the result of the
accumulation of materials on the membrane which no longer participate in the
mass-transport to or away from the membrane. In effect, they “blind” small sec-
tions of the membrane, thereby reducing the effective area and the flux through
the membrane.
I I I I I I I I , I I I
OO 40 80 120 160 200
TIME (days)
Figure 3.59: Long term flux decay and restoration by cleaning.
Often preventive measures may be taken to avoid fouling the membrane.

Prefilters or screens can be used to remove large particles which block thin chan-
nels or accumulate in stagnant areas of the module. High cross-flow velocities
tend to sweep deposits away. Low pressures avoid compaction of gels on the
membrane. Some polymers have a higher susceptibility to fouling and chemical
modification of the membrane surface can have a profound effect on the pro-
pensity to foul.
If fouling does occur, the membrane deposits can sometimes be removed by
aggressive cleaning agents, such as detergents, acids, bases, or even organic sol-
vents. The advantage of a chemically-resistant membrane is that severe cleaning
agents may be used. However, even with periodic cleaning, the flux cannot al-
ways be restored to the initial value. This results in an overall long-term decay.
Ultrafiltration 199
Effect of Cross-Flow Velocity

High cross-flow velocities tend to prevent fouling and also aid in the clean-
ing process. Figure 3.60 shows the flux decay on a log-log plot for low, medium
and high recirculation rates (QL, QM and QH). Plotting the data in this way of-
ten permits a reasonably good extrapolation of the flux for much longer times-
up to 2 or 3 years. The flux decay data usually plots on a straight line, and be-
cause of the cyclical nature of the log-log plot, 10 days of data often permits ex-
trapolation to 100 or even 1,000 days.
FLUX DECAY
J
Flux
10 100
TIME
Figure 3.60: Effect of cross-flow velocity on long term flux decay.
High cross-flow velocities also facilitate cleaning. Figure 3.614’ shows that
the flux is restored more rapidly and to a higher level with high velocities.
I
o-25
Cleanmg time (h)

The effect of detergent circulation rate
(-O- 1.51 l/s; -a- 1.14 I/s: -A- 0.57 I/s) and time
on flux during cleaning of tubular ultrafiltration
membranes.
Figure 3.61: Effect of recirculation rate on detergent cleaning.
Effect of Pressure
If no fouling occurs, the maximum flux will be obtained in the gel-polarized
regime above the threshold pressure (PT). However, for solutes which form semi-
gels on the membrane, pressures (PH) higher than the threshold pressure may
compact the gel layer resulting in greater fouling. Flux decay data like that of
Figure 3.62 may show greater flux stability at pressures (PL) lower than the
threshold pressure. Even though the initial flux at PL is lower (since not gel
polarized), the long-term flux at this pressure may be higher.
FLUX DECAY
J
Flux
10 100
TIME
Figure 3.62: Effect of pressure on long term flux decay.
Effect of Membrane Surface Treatments

Changes in cross-flow velocity or transmembrane pressure cannot always al-
leviate fouling. Membranes made from hydrophilic polymers like cellulose ace-
tate are generally less prone to fouling than the hydrophobic polymers. How-
ever, cellulose acetate is limited in its tolerance to high or low pH, organic sol-
vents and elevated temperature. In some cases, surface modification of the more
chemically-resistant polymers has rendered them less susceptible to fouling.
For example, a vinyl copolymer membrane (Romicon’s XM-50) was used
for years with anodic paint. The slight electronegativity of this membrane re-
pelled the negatively charged anodic paint producing stable flux over a long per-
iod of time. When the electropaint users switched to cathodic paints, the XM-50
membrane experienced catastrophic flux decline-presumably because of the
electrostatic interaction with the positively charged paint particles. Cellulose ace-
tate membranes provided a stable flux but could not be cleaned with the sol-
vents used in paint makeup (e.g., butyl Cellosolve). Therefore, a “charged” XM-
50 membrane (designated CXM) was developed for cathodic paints. Figure 3.63
compares the performance of the “charged” membranes with the uncharged.
Various functional groups may be applied to a membrane by techniques
such as chemical grafting, plasma polymerization, and sputtering. Figure 3.64
shows an increasing resistance of the protein gel-layer (I$) (decreasing flux) with
time for three samples of the same membrane, two which have been coated with
carbon and a polysiloxane. In this case, the inorganic coatings tend to resist foul-
ing producing a more stable flux.
Ultrafiltration 201
TIME-DAYS
Comparative data showing the flux versus time performance
of the XMSO membrane versus that of the newly developed
"charged" membranes on cathodic paint (PPG). The desig-
nation X1 and X2 refer to different "charge density" levels
Figure 3.63: Effect of “charged” membrane on long term flux decay with
cathodic paint.
R,.l
10-s bar msh
I
10.
uncoated
PSi -coated
C-coated
t
0 30 60 min
Figure 3.64: Effect of inorganic coatings on flux decay.
Howell and Velicangil 43I 44. have developed a “self-cleaning” membrane by

attaching food-grade proteases onto UF membranes. The enzyme protease hy-
drolyzed fouling proteins, thereby increasing the permeability of the gel layer.
This resulted in a 25 to 75% improvement in the permeate yield during a 22-hour

run concentrating 0.5% albumin or hemoglobin. Figure 3.65 shows the effect of
papain and fungal proteinase in decreasing the rate of flux decline over a 20.5-
hour run concentrating BSA from 0.5 to 5.0%. The net protein loss through the
membrane due to cleavange of albumin by the active enzyme was found to be
only 4% of the total. The enzyme may be crosslinked and immobilized on the
membrane surface with 0.125% glutaraldehyde in a phosphate buffer (pH 6.5).
Modelling flux of prototype and control/effect of different enzymes: PM-10 membrane,

BSA. (A) Control, Hl; (A) papain (Corolase SlOO), 11; (m) fungal proteinwe P, H2; (-1
model.
Figure 3.65: Effect of immobilized protease on UF of BSA (a self-cleaning
membrane).
MEMBRANE CONFIGURATION
The design of the membrane package or module and the fluid management
within that module will profoundly affect membrane performance. Further, the
optimum design for one application may be totally unsatisfactory for other ap-
plications. For example, Equations 17 and 22 show clearly that thin channels
promote higher mass transport and flux, but this must be balanced against their
greater propensity to foul. In addition, high performance modules must be eval-
uated in terms of cost, ease of cleaning and replacement. There are currently
four generic configurations for UF membranes in industrial use: tubes, hollow
fibers, plate and frame units, and spiral wound modules.
Tubes
Perhaps the simplest configuration is a tube with the membrane cast on the
Ultrafiltration 203
inside wall of the tube (see Figure 3.6). Tube diameters from 0.25 to 1 inch are
less prone to foul and more easily cleaned than any other configuration.
Tubes may be effectively cleaned by introducing a number of sponge rub-
ber balls slightly larger than the tube diameter (see Figure 3.66) into the process
stream. Periodic passage of these balls can prevent severe flux decline (see Fig-
ure 3.67). Figure 3.68 is a schematic diagram of a device to dispense and recover
these balls automatically at preset intervals. In this way, cleaning can be done
simultaneously with processing. When severe fouling is encountered, a cleaning
solution may be used in conjunction with the sponge balls.
The biggest disadvantage of tubes is their cost. The porous support tube (of-
ten fiberglass reinforced epoxy) is the dominant cost factor and the membrane
area per foot of tube is low. When the membrane is spent, usually the whole
tube must be replaced. PCI utilizes a “paper” insert on which the membrane is
cast as the replaceable element. Others have developed epoxy bonded consoli-
dated sand as the porous support in an attempt to reduce costs.
Removed deposit , Membrane supporter
Membrane Sponge ball
Figure 3.66: Schematic of sponge ball cleaning.
1 -
, - 0 With Sponge Ball Cleaning

l Without Sponge Ball Cleaning
1’ 2 4 6 8 lo 12 14 16 18 20 22
Operating Time (hr)
Figure 3.67: Effect of sponge ball cleaning on long term flux decay.
204 Handbookof Industrial MembraneTechnology
-Ball positioning handle

(Available in either manual
or automatic mode)
Permea.te
Pressure
Concentrate Feed
Figure 3.68: Sponge ball dispenser and collector.
The trend is toward using smaller diameter tubes or volume displacement

rods to reduce the volume of fluid pumped per unit area of membrane. For ex-
ample, the splined core in Figure 3.69 has sheet stock wrapped around the core
and sealed longitudinally. The tube is braided on the outside for strength. This
makes a fairly inexpensive tube, but the packing density (membrane area per
unit volume) is still quite high and there is considerable “parasite drag” due to
the splined core.
# CONCENTRATE
BRAIDED SUPPORT
(RETENTATE)
ULTRAFILTRATE
FEED
STREAM ULTRAFILTRATION
MEMBRANE
Figure 3.69: Splined core linear thin channel.

Ultrafiltration 205
Hollow Fibers
Conceptually, hollow fibers are the ideal membrane configuration. There is
no “parasite drag” and no expensive porous support tube. The fibers may be
pressurized on the inside (up to 30 psig) permitting “thin-channel” fluid man-
agement of the feed stream (refer to Figure 3.13).
The biggest disadvantage is the pressure constraint which limits the cross-
flow velocity down the lumen of the fiber. Shorter fiber lengths permit higher
velocities at maximum inlet pressures of 30 psig. However, since most of the
manufacturing costs are associated with potting the fibers at the ends of the
module (see Figure 3.70). shorter modules have a higher cost per unit of mem-
brane area. There are several methods under development for increasing the
burst-strength of hollow fibers to permit higher operating pressures.
Filtrate
hollow-fiber membrane
Hollow fiber terminal
1140mm
Figure 3.70: Schematic of hollow fiber module.
Potentially, hollow fibers should be the most economical membrane config-

uration available. Unfortunately, low manufacturing yields have kept selling
prices ($/sq ft) equal to or greater than that for spiral-wound modules. Reject
modules with only a few fiber leaks can be repaired using a “bubble point test”
(see Chapter 2). The open-ended module is immersed in a tank of water. A gas
line is used to pressurize the permeate port on the shell of the module with just
a few psig. The end of the module is observed to pin-point bubbling fibers (leak-
ers). These fibers are plugged using hot-melt, stainless steel nails with epoxy, etc.
The module is then turned upside down and the test repeated to plug the leaking
fibers on the other end. The same technique may be used in the field to reclaim
leaking modules.
In general, the hollow fiber configuration is more susceptible to fouling and
plugging than any of the other three configurations. Larger diameter fibers (up
to 2.5 mm I.D.) are becoming popular to improve fouling resistance. The lower
burst-pressures associated with these large diameter fibers generally limit the
maximum diameter to below 2 mm I.D.
Fortunately, hollow fibers may be cleaned by back-washing which tends to
compensate for their propensity to foul. Manufacturers of tubes, plate and frame
units, and spiral wound modules do not recommend back-washing due to prob-
lems with membrane delamination and glue line seal rupture. Because hollow fi-
bers are self-supporting and hold up well under the compression force of a re-
verse transmembrane pressure drop, they can easily withstand back-wash pres-
sures of 15 to 20 psi. However, the back-wash fluid should be filtered to remove
any particles which would tend to lodge in the porous wall of the fiber.
The permeate itself is often the ideal back-wash fluid since it has been
through the membrane once. Unfortunately, permeate by itself will not always
restore the flux (see Figure 3.7 1) and more aggressivecleaning solutions must be
used. An auxiliary pump and filter must be provided to deliver the cleaning solu-
tion under pressure through the permeate port to the shell side of the module.
The materials back-flushed will exit from the fiber lumen and must bedischarged
from the system so as not to redeposit on the walls of the fiber.
Water flux after backflushingwith NaOH
Initial water flux
Water flux after

bachwashing with water
Transmembrane pressure difference, lb/in’
Figure 3.71: Restoration of flux in hollow fibers by back washing. Cleaning

study showing the effect of back-flushing with sodium hydroxide on recovering
the flux of a cartridge intentionally fouled with cultured skim milk.
Perhaps the simplest cleaning technique of all is to continue circulating the

process stream through the module with the permeate port(s) closed (see Figure
3.72). The permeate will collect in the shell side casing of the module, until the
pressure builds up to the average of the inlet and exit pressures inside the fibers.
The inlet half of the module permeate will flow in the normal forward direction
through the membrane to the shell side; but in the exit half of the module, per-
meate will flow in the backward direction from the shell side to the inside (lu-
men) of the fibers. This is called “cleaning by recycling”. This gentle back-wash
of the exit half of the module often recovers up to 80% of the flux decay. The
reason is that hollow fibers often operate in laminar flow, and the gel layer is
most fully developed in the exit half of the module. Further, the sweeping ac-
tion of the process stream along the surface of the membrane effectively re-
moves material loosened by the gentle back-wash. If the modules are installed
with auxiliary piping so that the process stream flow direction may be reversed,
the other half of the module may be back-washed.
Ultrafiltration 207
Process return
Process in
Figure 3.72: Cleaning hollow fibers by recycling khutting Permeate Ports).

Plate and Frame Units

Flat sheet membranes in a plate and frame unit offer the greatest versatility
of any configuration but at the highest capital cost. In many of these units, vir-
tually any membrane may be cut to the appropriate shape and installed in the
unit. The membrane replacement costs are potentially low, but the replacement
labor is high. Disassembly also facilitates cleaning. There are many variations on
the plate and frame theme. Representative designs are shown in Figures 3.73 to
3.77.
One of the advantages of the DDS (Figure 3.73) and Rhone-Poulenc (Figure
3.74) designs is that each support plate has a separate permeate port. If a leak
develops in one membrane, it can often be located visually without disassembl-
ing the whole stack. Alternatively, the permeate tube from the leaking mem-
brane may be pinched off since the loss of one plate out of 50 to 100 plates is
not a significant percentage of the flow.
Figure 3.73: DDS plate and frame design.

Ultrafiltration 209
Figure 3.74: Rhone-Poulenc plate and frame design.
The Dorr Oliver design (Figures 3.75 and 3.76) heat seals the membrane to
plates which drain off filtrate through a central permeate port to the outside of
a pressure housing.
Millipore Corp. and New Brunswick Scientific Co. sell a "cassette" plate and
frame system. The "cassette" is a membrane packet (schematically shown in Fig-
ure 3.77) comprised of two membranes enclosing a filtrate collection screen
(sealed around the edges). Alternating holes on the edge of the cassette carry re-
tentate or filtrate to the appropriate manifold. The cassettes are separated by
screens or channel plates (see Figures 2.47 and 2.48 in Chapter 2). When filter-
ing cellular or colloidal suspensions, the channel plates are preferred over the
screens which collect particles on the cross members of the screen. Flat sheet
membranes with alternating gasket screens and channel plates may be used in
lieu of the cassettes only if the lateral permeability through the membrane is low
enough to prevent contamination of the filtrate with the retentate or leakage
out the edges of the stack. The maximum number of membranes in such a stack
is limited by the volume of flow which can be supplied through the manifold.
Since the membranes operate in parallel, the cross-flow velocity across each
membrane will diminish as the membrane area is increased.
Figure 3.75: Dorr-Oliver flat-plate membrane element.
Figure 3.76: Dorr-Oliver assembled flat-plate cartridge.

Ultrafiltration
Spiral Wound Modules

Spiral wound modules were originally developed for RO but are capturing
an ever increasing share of the OF market. They currently provide one of the
least expensive UF modules available in terms of cost per unit of membrane area.
A spiral wound module is essentially the cassette of Figure 3.77 rolled up in
a “jelly-roll” configuration. An envelope of two membranes enclosing a filtrate
carrier is sealed around three edges and the fourth edge is connected to a per-
forated tube which carries the permeate (product water) (see Figure 3.78). As
the module is rolled up, the membrane layers are separated by a screen or cor-
rugated spacer where the feed solution flows parallel to the tube axis (see Fig-
ure 3.79). Again, the corrugated spacers are preferable for cellular suspension
processing. The whole module is inserted in a pressure vessel; sometimes several
modules are placed in one long pressure vessel. Often a chevron seal is used to
seal the outer surface of the module to the inside of the pressure vessel; this in-
sures that all of the feed stream is forced between the membrane layers. How_
ever, for sanitary applications, a screen (controlled bypass spacer) is used to per-
mit flow in the annular region between the module and the pressure vessel; this
eliminates the stagnant area and facilitates in-place cleaning and sterilization.
PRODUCT WATER FLOW

(AFTER PASSAGE
THROUGH MEMBRANE 1
PRODUCT WATER
PROOUCT WATER SIDE BACKING

MATERIAL WITH MEMBRANE ON
EACH SIDE AND GLUED AROUND
EDGES AND TO CENTER TUBE.
,D.D04
MEMBRANE \
MATERIAL I
MEMBRANE
BRINE SIDE SPACER
Figure 3.78: Spiral wound module unrolled.

Controlled Anti-Telescoping Device
/ Bypass Spacer _ 7
Feed b-4 Concentrate
c
d% W
5--”
Permeate
Concentrate
Feed
Membrane
- Permeate Collecztion
c- Material
_>
w Membrane
Feed Channel
c....,,,
Figure 3.79: Spiral wound module showing screen spacer.

Spiral-wound modules cannot be unwrapped for cleaning lest the glue line
seal rupture; and most cannot be autoclaved. They are more prone to fouling than
tubes and some plate and frame units (depending on the type of feed channel
spacer), but they are more resistant to fouling than hollow fibers.
Table 3.3 compares the relative advantages and disadvantages of the four
basic configurations.
Table 3.3: UF Membrane Configurations
Tubular Hollow Plate & Spiral

Fiber Frame Wound
Cost/Area High Low High Low
Membrane High Moderate Low Moderate/

Replacement cost Low
(not includinglabor)
Flux (GSFD) Good Fair/Poor Excellent/ Good

Good
Packing Poor Excellent Good/Fair Good

Density (Ft2/ft3)
Hold-up Volume High Low Medium Medium
Energy High Low Medium Medium

Consumption
Fouling Excellent Poor Good/Fair Good/Fair
In Place Excellent Good Fair/Poor Fair/Poor
UF PLANT DESIGN
Mode of Operation
The arrangement of the membrane modules and their mode of operation
can affect the economics as much as the module design. There are basically 3 dif-
ferent operating modes for UF. These are shown in Figure 3.80.
Single-pass UF can only be used on relatively pure water streams (e.g., de-
ionized water). If the retained species are low in concentration, the “recovery”
of permeate can be as much as 95 to 99% of the feed. In the case of deionized
water, the colloidal silica, microorganisms, and organic molecules in the concen-
trate can be sent to the drain as “blow-down”. In the case of the hundred-fold
concentration of a dilute hormone, 99% of the feed will pass through the mem-
brane. Although the single pass operation in Figure 3.80 shows the UF modules
operating in parallel, it is sometimes advantageous to operate in series to main-
tain a longer contact time with the membranes and to reduce the volumetric
Ultrafiltration 215
pumping requirement. The plate and frame units can often be arranged to op-
erate with the membranes in parallel or in series.
In most cases, the flux is too low to operate in the single-pass mode; the re-
covery of permeate in a single pass is a small percentage of the feed (very little
concentration of retained species). Recirculation of the process stream across
the membrane is necessary to obtain the desired concentration or recovery. Typ
ically the ratio of recirculation rate to permeate flux is 10 to IOO-fold. Recircu-
lation can be accomplished with either a batch concentration or with a feed and
bleed operating mode.
*
RMEATE
A. BATCH ULTRAFILTRATION
(;LJr l BLEED
FEED
*
B. FEED AND BLEED ULTRAFILTRATION
C. SINGLE-PASS ULTRAFILTRATION
Figure 3.80: UF operating modes.

In a batch operating mode, the retentate is circulated back to the feed res-
ervoir. As the permeate is removed, the volume in the reservoir goes down and
the concentration of retained species goes up. Freely permeable species (e.g.,
salts) remain at the same concentration in the reservoir and in the permeate
stream. Eventually, the volume in the reservoir becomes too small to pump and
the run is over. A new batch must be charged to the reservoir to continue.
The feed and bleed mode permits continuous filtration; the feed stream is
fed into a recirculating loop. The concentration of retained species will continue
to increase with ever decreasing flux, unless a purge stream is taken from the
loop.
Often a ratio controller is used to keep the feed to bleed ratio constant-
equal to the volumetric concentration ratio required. This means that even with
flux decay, the concentration ratio and recovery will remain constant.
It is unwise to use the feed and bleed mode to concentrate a batch of solu-
tion. Once steady state is achieved, the membrane filter at the final concentra-
tion for the duration of the run. In a batch process, a higher average flux is
achieved due to the gradually increasing concentration. Thermodynamically,
the feed and bleed mode is less efficient due to increase in entropy upon mixing
the low concentration feed with the high concentration recirculating retentate.
The total membrane area required in a feed and bleed mode can be reduced
considerably by arranging the modules in two or more feed and bleed stages. In
this multistage (cascade) system, the bleed stream of the first stage becomes the
feed stream of the second stage, etc. For example, the first stage might concen-
trate from 1 to 2%, the second stage from 2 to 4%, the third stage from 4 to 7%,
and the last stage from 7 to 10%. Since flux decreases with increasing concen-
tration, it is advantageous to operate as many stages as possible at the lower con-
centrations. Obviously, the additional pumps and controllers for each stage add
to the capital cost and must be considered in arriving at an optimum number of
stages. (See Chapter 6: Process Design and Optimization.) Though the optimum
number of modules per stage will vary, equal numbers of modules for each stage
often comes close to the optimum configuration.
Optimum Recirculation Rate

There is an optimum recirculation rate for each stage. Higher recirculation
rates will result in greater productivity (flux), less membrane area, and longer
membrane life due to greater flux stability. They also require larger pumps and
increased power costs.
Figure 3.81 shows some of the major operating costs. Increasing the cross-
flow rate through a module decreases the membrane replacement costs due to
less area (higher flux) and longer life (greater flux stability). Indeed, all costs as-
sociated with the size of the plant will go down, but the utilities costs go up. The
optimum design point will depend on the current cost of power and membranes
(along with their productivity).
In a multistage feed and bleed system, the optimum cartridge flow rate can
be different for the later stages operating at high concentration than those oper-
ating at low concentration. Often, the first stage is not fully gel polarized (due to
the low concentration) and high cross-flow velocities are not as critical as high
operating pressures (approaching the threshold pressure).
Ultrafiltration 217
oo-
Labor t supervision t maintenance
I I I 1 1
O'O 20 40 60 60 100
Cartridge flaw rate, gallmin
Figure3.81: Determination of optimum recirculation rate.
APPLICATIONS
The largest current applications for UF (in terms of installed membrane area)
were not envisioned in the early sixties (e.g., UF of electrocoat paint). Further,
many of the projected applications have not yet materialized (e.g., fractionation
of polymers and proteins).
In some applications, the product is the retentate, and the objective is to con-
centrate or purify the retained species (passing unwanted contaminants through
the membrane).
In other applications, the product is the permeate, and the objective is to re-
move unwanted contaminants which are large enough to be retained by the

membrane.
In a few applications, both retentate and filtrate are important. For exam-
ple, if a valuable product or by-product is a pollutant in a waste stream, recovery
and use of the product will often pay for’the pollution abatement. The clean
permeate may then be reused in the plant; in caseswhere the plant effluent is at
elevated temperatures, the hot permeate may be reused in the plant saving addi-
tional energy costs. There are only a few applications where UF can be justified
for pollution abatement without the recovery of a valuable by-product or energy
credit. In these cases, the economic incentive is found in the avoidance of a
sewer tax or shut down by the Environmental Protection Agency.
Ultrapure Water
Semiconductor Industry. In Chapter 2, the application of MF and RO in a
deionized water loop was discussed. The rapid miniaturization of integrated cir-
cuits (I&), and the need for even greater water purity has prompted the use of
UF.
Traditionally, MF has been used after the ion exchange columns (see Figures
2.56 and 2.58) to remove microorganisms, resin fines and other particles. Increas-
ingly, UF is being used after the ion exchange units and before MF. Originally,
UF was justified on the basis of savings in the MF replacement cost since reduc-
ing the load on the MF cartridges naturally extends their life. In one case, the
savings in replacement cartridges and labor were expected to pay for the U F plant
in less than a year.
Except for ultrapure water plants where the DI water is dirty and the MF
cartridges must be changed every two weeks or less, MF operating costs are con-
siderably less than UF. Typically, MF cartridges can last from 3 to 6 months
(even without UF). Even with replacement every other week, MF operating costs
are usually no more than 509 per 1,000 gallons. UF operating costs often run
more than this when membrane replacement costs, labor, maintenance, power
usage, and amortization are included.
The economic justification for UF is usually found in improved IC yields
due to better quality water. An 80,000 MWCO UF membrane can remove all
particles and macromolecules larger than 0.005 /..I, and a 10,000 MWCO can re-
move particles down to 0.002 cc.This means that high molecular weight organics
and colloidal particles, which pass through an MF membrane, will be retained by
UF. The organics adhere to the silicon chips resulting in poor adhesion of the
polymer masks used to control etching. Further, dead bacteria fragments/pyro-
gens (down to 10,000 MW) can adhere to wafer surfaces resulting in MOS gate
failures. “Pin-holing” in angstrom thick oxides can occur due to pyrogens. Re-
cent reports (DM Data Inc., Scottsdale, Arizona) indicate that Japanese manu-
facturers of silicon chips get almost twice the yield of usable chips that U.S.
manufacturers get. One significant difference between U.S. and Japanese semi-
conductor plants is that most of the large Japanese plants are using UF. Cur-
rently, only a few U.S. plants have UF.
The effectiveness of UF in particle/colloid filtration can be measured with
an MF plugging test similar to the “silt density index” (see Chapter 4, Figure
4.10). Water is fed at 30 psig through a 47 mm diameter 0.22 p membrane filter
Ultrafiltration 219
monitoring the flow rate with time. Figure 3.824’ shows the improvement in
the quality of the UF permeate at the Western Electric plant in Allentown, PA.
It should be realized that the water is relatively clean (the UF unit is operating
in the single-pass mode with 95% recovery), and that the use of a 47 mm filter
accelerates the time to plugging.
UF PLUO TEST
FILTER - 0.22 llIC2ONS
loo
I 5 10 50 100
TIIIE (HOURS)
Figure 3.82: Effectiveness of UF (ultrapure water) using MF-plugging test.
The removal of organics is measured as a reduction in total organic carbon

(TOC). With an 80,000 MWCO membrane, there is usually no problem in meet-
ing the specification cited in Table 2.7 (Chapter 2) of less than 0.2 ppm TOC.
In some case, a 20,000 or 10,000 MWCO membrane will be necessary to meet
the more stringent Japanese specification of less than 0.05 ppm TOC.
Though the retention of microorganisms by a UF membrane is well over
99%, it is not as “absolute” as an MF membrane (see Chapter 2). For this rea-
son, it is wise to install an MF cartridge at the point-of-use for final insurance.
(Replacement should be infrequent.) The Center for Disease Control has re-
portedM that spiral-wound modules produce water with consistently lower bac-
teria counts than hollow-fiber modules when challenged with high populations
of bacteria in the feed water. With continued filtration, bacteria counts on the
membrane increase exponentially. For this reason, all UF membranes must be
sanitized periodically. Sodium hypochlorite is preferred, but cannot be used
with polyamide membranes. It also attacks stainless steel. Table 3.4 lists com-
mon sanitizing agents.
Table 3.4: Common Sanitizing Agents *
Hz02 I-S% hydrogen peroxide for 30-90 minUteS
(preferred in electronics where sodium iOn

residuals must be avoided.)
C’2 100-500 mg chlorine gas per liter of water for

30-90 mlnutes.
NaOCl S-10 ppm sodium hypochlorlte for 60 mlnUteS.
Hot H20 IBO’F for l-2 hOUrS
03 Ozone at 0.1-0.2 mg/liter for 30-90 minutes
* Chf~k for chemical cumpetibllity with the filters usedin the System before selectfng 8
sanitizing agent
UF sometimes provides an unexpected bonus in reducing the conductivity

of the water. Though “charged” UF membranes with a low MWCO might offer
some rejection of divalent and trivalent ions, they should pass an 80,000 MWCO
membrane. Figure 3.8345 shows tremendous fluctuations in the influent conduc-
tivity. UF reduced the conductivity to a stable minimum. It is hypothesized that
charged colloidal particles or large molecular weight complexes contribute to the
conductivity.
UF CONOUCTIVITY VS. TIME
LNFLUENT UF
.05
2 4 6 8 10 12 14 16 18 20
DAYS
Figure 3.83: Stabilization of ultrapure water conductivity with UF.

Ultrafiltration 221
A more recent development is the use of UF for point-of-use processing.

When UF is installed in the central syste’m, the plastic (PVC) piping to the vari-
ous rinse stations collects particles and leaches organics. A small 4 gpm hollow
fiber unit at the rinse station removes all particles, colloids and organics down to
an average size of 0.005 ~1, In some cases, these units are operated dead-ended
with a periodic fast-forward flush to remove accumulated particles inside the
hollow fibers.
UF has also been used in ultrapure water loops before the ion exchanger
units to protect resins from fouling and as pretreatment for RO.
Pharmaceutical Industry. Water is, of course, the most common ingredient
in pharmaceutical products. It is crucial that all microorganisms, pyrogens, and
other particles be removed from water for injection (WFI). The smallest blood
vessels in the body have a diameter slightly less than 3 p. Thus, particles larger
than 3 &r can block blood vessels resulting in partial occlusion of retina arteries
(leaving blind spots), clot formation and emboli, and granulomas. Though MF
membranes can remove the microorganisms and particles above 0.2 /..&they can-
not effectively remove pyrogens. It has been repeatedly demonstrated that a
10,000 MWCO UF membrane can remove all pyrogens.
The word “pyrogen” means “fever-producing” and has been used to cover
any substance which causes a body temperature increase on injection. In fact,
the first compendia1 pyrogen test was published in the 1942 edition of the
United States Pharmacopeia. It involved the measurement of the rise of body
temperature in rabbits upon intravenous injection of a test product. More re-
cently (1977), the FDA approved the Limulus amebocyte lysate (LAL) test
which can be run for less than 10% of the cost of a rabbit test.
Because of their fever-producing capability, pyrogens must be barred en-
trance to the blood stream, particularly in persons who are ill or whose immune
systems are weaker than normal. It is well known that pyrogens can produce
shock and even death.47
Historically, pyrogens have been defined on the basis of their physiological
action rather than on their chemical characteristics. However, in recent years
considerable progress4’ has been made in identifying pyrogens as lipolysaccha-
rides (LPSj primarily from Gram-negative bacteria (GNB). These materials occur
in the outermost layer of the cell walls; thus, a pyrogenic response can result
from the presence of dead bacteria or even cellular debris. This explains why
normal autoclaving does not destroy pyrogens and filtration through 0.2 p steri-
lizing membranes does not remove them. Further, since GNB grow in solutions
containing minimum nutrients, even distilled water can evoke a pyrogenic re-
sponse. Indeed, pyrogens can even be entrained and carried along with water
droplets into the condenser during evaporation.
Because the LPS molecules contain both hydrophobic and hydrophilic re-
gions, they form micelle-like aggregates. If the LPS is reduced to its smallest
subunit, it will pass through a 1,000,000 MWCO ultrafilter but be retained by
a 10,000 MWCO membrane. 47 Experiments show that the 10,000 MWCO mem-
brane will reduce the pyrogen concentration by 2 to 5 orders of magnitude.4a
Other researchers4’ seeking to remove pyrogens from peritoneal dialysis solu-
tions found that a 50,000 MWCO membrane was adequate and provided a higher
flux.
Electrocoat Paint
The electrocoat paint market constitutes the largest single application of UF
in the world. Virtually every automobile plant in the world uses electrocoat for
the undercoat and virtually every installation utilizes UF. There are over 1,000
UF units in the automotive industry alone-not to mention appliance manufac-
turing, metal furniture, coil coating, and other metal electrocoat operations.
The electrocoat process for primer coating involves the electrophoretic
deposition of charged colloidal resinous particles in aqueous dispersion onto
a conductive substrate such as an automobile body. The process is universally
favored because once the paint particles are deposited, they insulate the body
from further deposition at that point. The impressed electric field thereby causes
the migration of paint particles to uncoated areas. The result is an extremely uni-
form, coherent and defect free coating-even on sharp edges and in recessed
areas inaccessible to other methods.
UF provides a cost effective way of solving several problems associated with
electrodeposition systems. The continuous feed of paint to the tank is balanced
only by the deposition of charged particles on the substrate. This results in the
accumulation of several water-soluble ions in solution with consequent decreased
“throwing power”, lower rupture voltage, staining, thin films, and pinholing due
to rupture. Further, excess drag-out paint from the paint tank must be rinsed off
the metal part to prevent “orange-peel” and other coating anomalies. This dilute
paint cannot be returned to the tank due to the excess water. It is, therefore, dis-
carded to the sewer resulting in a horrendous pollution problem and tremendous
lossesof paint.
At first, in the early nineteen-sixties, UF was used as a kidney to remove
contaminating ions. Later, PPG (a paint manufacturer), envisioned UF in a closed
loop rinse system (see Figure 3.84) to solve the pollution problem as well as the
contaminant build-up. In this system, a portion of the ultrafiltrate is purged to
the drain to remove ionic contaminants. The majority of the permeate is used in
the rinse tunnel to wash excess drag-out paint back into the tank. Thus, all ex-
cess paint is recovered alleviating a serious pollution problem and, at the same
time, justifying the UF unit economically.
Rinse
I I
Paint makeup
i
To dr,oin
Figure 3.84: UF of electrocoat paint, single-rinse system.

Ultrafiltration 223
The size of the UF unit may be reduced considerably by utilizing a counter-

current rinse scheme (Figure 3.85). The ultrafiltrate is used only at the last rinse
station. The dilute paint (0.05 to 0.2%) from this rinse is used in the first and
second rinse stations, eventually returning all excess paint to the paint tank.
The paint savings alone typically pay for the UF unit is less than six months.
Additional savings result from reduced deionized water use, lowered waste treat-
ment costs, and better control of bath composition.
Optional
1st rinse 2d rinse 3d rinse DI rinse
-_
To d&n
Figure 3.85: UF of electrocoat paint, multi-rinse system.
In the nineteen-sixties, most electrocoat tanks were anodic, i.e., the work
piece was positively charged and served as the anode while the paint particles
were negatively charged. These systems were ideal since many UF membranes
tend to be slightly electronegative with good fouling resistance and excellent
flux stability (Figure 3.86).”
Figure 3.86: UF of electrocoat paint-stability of anodic paint flux with XM-50

membrane.
In the nineteen-seventies, a cathodic paint process was developed where the

work piece was negatively charged and the paint particles positively charged.
This was a superior coating process but led to severe fouling problems with some
of the standard UF membranes (Figure 3.63).50 The advantage of the cathodic
process is related to the fact that there is no metal oxidation and hence no metal
ions in the paint film, giving improved resistance of the finished article to corro-
sion. The membrane fouling problem was solved by using positively charged
membranes as in Figure 3.87 or membranes with less electronegativity.
--*---- SHUTDOWN PERIOD
2
III, III,, II II I ,,,,,r,,,,,,,,,,,,,,
2 4 6 8 IO I2 I4 10 IO 20222420 a So22 243628 4042444643 805234~6~606264666~
TIME (DAYS)
Figure 3.87: UF of electrocoat paint-stability of cathodic paint with CXM

membrane (positively charged).
Oil-Water Separations
Billions of gallons of oily wastewater are generated daily. Strict environmen-
tal legislation now requires industry to clean up these wastes.
Industrial oily wastewaters can be divided into three broad categories ac-
cording to the distribution of the oil phase:
(1) Free-floating oil

(2) Unstable oil-water emulsions
(3) Stable oil-water emulsions
Free oil is readily removed by mechanical gravity separation devices. Un-

stable oil-water emulsions can be broken mechanically or chemically. It is the
stable oil-water emulsions which are most difficult and at the same time most
.amenable to treatment by UF. Chemical coagulation/flotation or contract haul-
ing are usually more expensive alternatives. Further, all chemical treatment meth-
ods produce a sludge in which the dirt, floe and trapped water remain in the oil
phase.
Stable oil-water emulsions are generated in many diverse industries. /Vera/
working operations use water-soluble coolants, cutting and grinding oils, and Iu-
bricants for machining. Metal cleaning tanks and alkaline degreasing baths gener-
Ultrafiltration 225
ate an oily wastewater. Rolling and drawing operations use oil lubricants and
coolants. Food processing has waste streams with natural fats and oils from ani-
mal and plant processing-particularly vegetable oily wastes. Peel oil is another
example from the citrus industry. Textile manufacturing will produce natural
oils from wool scouring or fabric finishing oils.
Metal cleaning and wool scouring wastes are illustrative of the diversity of
oily wastewater treatable by UF.
Metal cleaning operations normally precede painting or plating operations-
the objective being to remove dirt and grease. UF can be very useful in this ap-
plication in extending the useful life of the wash water and reducing the waste
disposal problem. Figure 3.88 is a schematic of a prepaint phosphating line be-
fore and after introduction of UF. The phosphating treatment prepares the metal
surface for bonding with the paint. A well run phosphate line counterflows the
rinse stages into the first stage cleaner tank; fresh water is introduced into the
third stage rinse which overflows into the second stage, etc. The buildup of oil
and dirt in the alkaline cleaning stage is removed by UF; the hot clear, detergent-
laden washer solution is then returned for reuse in the first stage reservoir. This
closedcycle counterflow system lowers the oil level in the third stage, reduces
detergent costs, and decreases spray nozzle clean-out maintenance time. Further,
the savings in disposal charges alone (30-fold reduction in volume) can pay for
the UF system in less than two years.
BEFORE
Work Flow Make-Up
Make-Up Make-Up Make-Up
b
I I I I
mm A +A A +A A +A A +A AA
Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 Stage 6

Detergent Water Water Phosphating Water Neutralizer
Wash Rinse \ Rinse \ Rinse \
Drain $ Drain a
40,OOD GPD !?hO
, G: ~~ 30,000 GPD
AFTER Make-Up Make-Up Make-Up

I I I
AA.AA A+A A +A A +A AA
+-- Make- +.- Make-

UP UP
Stage1 2 Stage 2 Stage 3 Stage 4 Stage 5 \ Stage 6
Ultrafiltrate x X
Feed Drain ’
Drains Closed Off 30,000 GPD
Concentrate
1c
Ultrafiltration
System
Figure 3.88: Flow schematic of prepaint phosphating line with and without UF.
In the wool scouring process, aqueous detergent solutions are used to remove
contaminants from raw wool-principally wool grease and suint (sheep sweat)
with smaller amounts of soil and fecal matter. During scouring, the wool grease
is emulsified by the detergent solution, the suint dissolves, and the mineral par-
ticles become suspended. The waste stream is highly polluting with a COD of
80,000 ppm. UF has been effective in reducing the COD to less than 15,000
ppm as well as decreasing disposal costs.
Oily waste waters suitable for treatment by UF contain 0.1 to 10% oil in a
stable emulsion. A limited amount of free oil can be processed but usually quan-
tities above 1 to 5% are removed with a centrifuge prior to UF. The difficulty
with free oil or unstable emulsions is that the oil accumulates at the membrane
interface and may form a continuous layer which preferentially wets the mem-
brane over water (the surface tension is lower). In this case, the membrane will
pass oil and retain water. (See Chapter 2 on the bubble point test). The secret
of successful UF is to maintain discrete and stable emulsoid particles of oil (gen-
erally over 0.1 /..rin size) which are larger than the membrane pore size (0.01 /.I
or below). When this is the case, oil in the permeate will generally be less than 10
to 50 ppm.
In the presence of poorly stabilized oil emulsions, cellulosic membranes sig-
nificantly outperform medium surface free energy membranes which, in turn,
significantly outperform low surface free energy membranes. Noncellulosic
membranes can become wetted with oil more easily-losing their water flux and
rejection for oil. If the chemical nature of the oily wastewater requires a more
chemically resistant hydrophobic membrane, it is important that a high cross-
flow velocity be maintained. (Oil wetting of hydrophobic membranes begins at
stagnation points.1
Oil-water emulsions behave like colloidal suspensions showing typical “gel-
concentrations” of 75 to 80% (close packing of spheres) and a flux independent
of pressure (in the “gel-polarized regime”). Low lubricity emulsions (with a high
“synthetic” oil content) are usually characterized by high flux and low fouling
rates. Oils containing large amounts of “tramp oil” with moderate to low lubric-
ity show lower flux and require more frequent membrane cleaning. High lubric-
ity oils (containing natural fatty materials) exhibit low flux and are prone to
foul the membrane.
Cleaning is normally done with detergent solutions at elevated tempera-
tures (60°C) and high pH for 1 to 2 hours. Usually, UF can concentrate the oil
up to 40 to 60%. This can reduce the contract hauling costs by two orders of
magnitude. Further, oil concentrations of over 50% can support combustion. If
the oil concentrate can be burned in a boiler, any fuel credit can help offset the
cost of the UF plant.
Reclamation of Waste Lubricating Oil

U.S. motorists discard about 1.25 billion gallons of oil every year by chang-
ing the oil in their automobile crankcase. Despite the fact that these are among
the highest quality products refined from our best crude oils, only about 100
million gallons reach recyclers to be re-refined into clean lubricants. The rest is
burned as fuel or disposed of as waste. Indeed, a conservative estimate is that
over 500 million gallons of used lubricating oil are annually injected directly
into the environment through landfills, burning, and other disposal methods.
Ultrafiltration 227
The idea of recycling these lubricants has flourished when oil is scarce. Pres-
ently, the oil glut has put many re-refiners out of business. In 1935, Adolph
Hitler prodded German industry to collect and recycle its dirty lubricating oil
to help reduce oil imports. Since then, subsidies and legislation have made the
concept a permanent fixture in Germany. It is said that even today some two-
thirds of West Germany’s waste crankcase oil is recycled. Eventual shortages of
petroleum reserveswill necessitate some form of recycling.
The standard re-refining process uses sulfuric acid to dissolve sludge. The
oil phase is further cleaned by clay filtration. The process is so inefficient that
one-third of theoil is lost and disposal of the acid solution is increasingly difficult.
The “water-alcohol“ method utilizes isopropyl alcohol to decompose metal-
lic soaps. The mixture is then centrifuged to yield clean oil and a watery metallic
sludge. Alcohol is recovered for reuse.
UF has the potential of removing all contaminants, but the flux is very low
at room temperature. Inorganic UF membranes operating at 300°C and a pres-
sure of 7 bar (105 psi) are capable of processing the oil economically. In one
plant in Europe, where the spent lubricating oil is pretreated with thermal shock
and centrifugation at 18O”C, the UF flux is reported stable between 1000-2000
LSMD (25-50 GSFL) over six months without cleaning.
Decontamination of Crude Oil

Nearly all metal contained in crude oil is either chemically bound with as-
phaltene type structures or associated with asphaltenes (IOOO-10,000 daltons).
It is necessary to remove these metals to prevent contamination of catalysts.
The present process uses a solvent like hexane or pentane to flocculate the
asphaltenes and is relatively expensive. UF can retain the metal asphaltene com-
plexes while passing virtually all other crude oil fractions. Again, the flux is low
due to high viscosity. However, an inorganic membrane operating at 330°C on a
10% asphaltene feed produced a permeate with less than 0.5% asphaltenes.
The removal of the carbon-forming asphaltenes and catalyst-fouling metals
also reduces the process costs for desulfurization of fuel oil.
PUA Recovery
Polyvinyl alcohol (PUA) is used as a sizing agent to improve the strength
and surface characteristics of warp yarns prior to weaving operations, where they
are subjected to considerable abrasion and tensile stress. The traditional warp
sizing agents (starch, gelatins and gums) have been replaced by improved syn-
thetic materials like PUA or sodium carboxymethylcellulose (CMC). PUA sales
for this application now exceed 40 million pounds annually. Later in the textile
manufacturing process, the sizing agent is removed from the cloth by scouring
before dyeing and finishing operations. Because the synthetic sizing agents are
not biodegradable, they pose a serious pollution problem.
Although chemical precipitation of these sizing agents is possible, there is a
sludge disposal problem. Due to the high cost of PUA and CMC, their recovery
with UF and reuse is a much more attractive solution. Figure 3.89 is a schematic
of the PUA recovery system. Not only can the PUA be reused, but hot permeate
(185’F) can be recycled directly to the desizing bath. The dollar value of the
PUA and energy recovered will generally pay for the UF system in 10 to 18
months.
further processing
,20-50 pm cartridge filter
I
3ving
ther
i
-Makeup
Permeate return
ratio ins
-
water
I
Ultrafiltration sys rm
I
LP
Balance
-_ -..
_..-..
- _. -
-..
__
-
_.- ..
tank
em-.. -.. _
-..-.. _. .
-.. -.. _
-..-.. -..
Slasher
Mixing tank
Figure 3.89: Flow schematic of PVA recovery system.
Ultrafiltration 229
The UF membranes are protected from yarn fibers by a vibrating screen

filter and 20 to 50 /J cartridge filters. This makes possible the use of spiral
wound modules which have a life of 24 to 30 months. The desizer waste ef-
fluent usually contains between 0.5 and 1.5% PVA. The UF concentrates this
PVA up to 10% for direct use in the slasher. This final concentration is monitored
and automatically controlled by an in-line refractometer. A small amount of de-
sizer waste is purged to drain to prevent the buildup of low MW solutes.
Dyestuff Recovery and Purification

Many dyes are too small to be retained by a UF membrane. Exceptions in-
clude polymeric dyes and indigo; the latter has a low molecular weight (262 dal-
tons) but can be retained by a 50,000 MWCO membrane, when in the oxidized
state due to its insolubility in water.
The economics of an indigo dye house can be improved dramatically with
the recovery of indigo from the plant effluent. Losses through the rinse system
can account for more than 10% of the mills’ total indigo consumption. A mem-
brane life of 3 years is reported.
Whether recovered by UF or RO, it has been established that recovered dyes
can be reused in dyeing operations with no attendant problems and meeting all
color specifications.‘l
UF has also been evaluateds2 as a method for purification of polymeric dyes
made by the attachment of an azo-chromophore onto a polymeric backbone.
This is particularly important when these dyes are used as food colorants. The
low molecular weight species and oligomers must be removed to ensure that the
product will be nonabsorbable following ingestion.
As expected, these polymeric dyes were rejected completely by the mem-
brane. Surprisingly, Sunset Yellow (MW 452), Schaeffer’s salt (MW 246) and tar-
trazine (MW 534) chromophores were significantly rejected by membranes with
MWCO’s in the range of 10,000 to 50,000. It is known that dye molecules of
this type can associate in solution; but an increase in NaCl concentration, ex-
pected to increase the degree of association, decreased the rejection of the dyes.
Therefore, it is more probable that the dye molecules adsorb to the membrane
making an effective anionic membrane rejecting the charged dyes. At high ionic
strength, the charge on the membrane is screened allowing passage. It was found
by Cooper et aIs that a sodium chloride concentration of 15 g/2 was necessary
to achieve good transport of the small MW chromophores. Alternatively, a high
pH’? had a similar effect.
Latex Concentration/Recovery
The concentration of latex emulsions was one of the early applications sug-
gested for UF. Unfortunately, not all latexes are amenable to processing with
UF. Many latexes are unstable under the high shear induced by pumps or even in
the thin channels of UF modules. However, the use of diaphragm pumps and
careful control of hydrodynamic shear within the module helps prevent coagula-
tion.
UF has been used in place of evaporation to concentrate in-process latex
streams from 30% to as high as 65% with reduced energy consumption. Poly-
vinyl chloride is the most prominent latex in this application.
UF has also been used to clean up polymerization kettle wash waters before
disposal. The dilute latex can be concentrated from 0.5 to 25%, thus reducing
the volume to be hauled away to ~/SOof the original volume. In some cases, the
waste latex is recycled for reformulation. Where there is a significant sewer tax,
UF is an economical alternative even without recovery of the latex. The dilution
of latex during polymerization kettle rinse-down may deplete the surfactant on
the polymer particles or introduce multivalent ions, resulting in decreased latex
stability. Adjustment of pH or addition of surfactant can prevent coagulation of
the latex on the membrane.
If the feed is preconditioned properly, the UF flux is often quite stable. One
thousand hours of continuous operation between cleanings is common, When
flux decay does occur, detergent washing is usually sufficient to restore flux. In
some cases, polymer solvents may be required. Proper selection of a solvent re-
sistant membrane and/or solvents which will dissolve the latex but not affect the
membrane is crucial. For PVC latex, the solvents of choice are methyl isobutyl
ketone (MIBK) and methyl ethyl ketone (MEK). Styrene butadiene rubber will
swell in MIBK, MEK or toluene. Polyvinyl acetate will dissolve in the low MW
alcohols such as propyl alcohol. Generally, the membranes are first washed with
water, then detergent, followed by another water flush. The system is then
drained of all water since it will affect polymer solubility in MEK. Finally, a
solvent rinse is employed. If the module is tubular, sponge balls will enhance
cleaning.
Removal of Heavy Metals

UF can remove metal values from metal plating wastes provided the metals
are first precipitated to form a colloidal suspension. The process is described in
Chapter 2 in the section on cross-flow applications. Even though MF membranes
provide a higher flux, they are more prone to flux decay than UF due to internal
pore fouling. The long term UF flux may be higher than that from MF.
Pulp and Paper Waste Treatment

Although UF cannot recover the low MW sugars in waste from the pulp and
paper industry, it can remove lignin compounds and most of the color bodies.
Pulp is made by separating wood into individual fibers. This is done mechan-
ically or chemically. Only 50% of the material in the tree is used as pulp. The re-
mainder is often discarded as waste in the spent pulping liquor.
In the mechanical process, logs are pressed against a rotating grindstone; the
resulting fibers are mixed with water and screen filtered. This mechanical pulp
contains lignin, gums and mineral salts in addition to the wood fiber.
In the chemical pulp process, wood chips are digested with a chemical
“cooking liquor” under high pressure and temperature. There are three major
cooking processes. The sulfate or kraft process is the. most prominent; about
75% of the wood pulp produced in the U.S. is made by the kraft process. It uses
a solution of NaOH and Na2S, called “white liquor,” to remove the lignin that
binds the cellulose fibers together in the wood. The loose fibers are then sep-
arated from the spent cooking solution-called “black liquor”. The soda process
uses caustic (NaOH) and the sulfite process uses sulfurous acid (HsSO,). If the
pulp is to be used in high-grade white paper, it is bleached with chlorine.
Ultrafiltration 231
Much of the BOD contribution is pulping wastes comes from low MW car-
bohydrates. RO can often reduce the BOD by 70 to 90% whereas UF effects
only a 45 to 55% reduction in BOD. 53 However, UF can remove 85 to 98% of
the color bodies.
UF is finding some utility in recovering lignin compounds-used in the pro-
duction of vanillin, adhesives, detergents, and dispersants. Separating the sugars
and lignin compounds solves the problem of sugar contamination in the feed
stream to a vanillin plant. Unfortunately, a clean fractionation between the two
is not possible even with UF since some of the sugars are bound in lignin-carbo-
hydrate complexes.
Table 3.5% shows the retention of sodium base spent sulfite liquor by a ser-
ies of UF membranes. The multistage UF was carried out in succession from the
largest to the smallest pore size-i.e., the permeate from the first stage became
the feed for the second stage. Although 90% of the sugar passed through mem-
branes between 100,000 and 20,000 MWCO, the retention increases for 10,000
and 500 MWCO membranes up to 36%. It has been estimated that about ‘16of
the sugars are bound in complexes with molecular weights above 10,000 and an
additional 10% in complexes above 500 daltons. In other words, ‘/4 of all the
monosaccharides present in spent sulfite liquor are bound. On the other hand,
100,000 MWCO membranes retained over 50% of the lignosulfonates. It appears
(from Table 3.5)% that about 70% of the lignosulfonate compounds have a MW
over 50,000; only 7.2% had a MW less than 10,000.
Table 3.5: UF of Sodium Base Spent Sulfite Liquor-Retention of Sugars 81

Lignins
Original Original
sugar, % lignin. %
Nomlnal reten-
Ultrafilter tlon, mol.wt. Passing Retained Passing Retained
It has also been shown that hydrolysis of the spent sulfite liquor with 6%
H3P04 can improve lignin retention by raising its MW through polymerisation.s4
The discharge of colored effluent is regulated primarily for esthetic reasons,
but it may also interfere with plant and animal life cycles by blocking sunlight.
Color bodies from pulp and paper mills are difficult to remove by conventional
sedimentation and biological treatment; it requires massive lime treatment and is
ineffective with spent sulfite liquor (less than 80% color removal). UF can usu-
ally remove over 90% of the color bodies and often achieves over 98% removal.
The rejection data of Table 3.6” indicate that more than 97% of the color-
bearing material has an effective MW greater than 10,000. In these data, the pre-
meate from the UM 10 membrane was fed to the UM 2 membrane and its per=
meate was fed to the UM05 membrane. Thus, the UM2 membrane can remove
99% of the color, 64% of the COD and 43% of the TDS.
Table 3.6: UF of Caustic Extraction Effluent-Color Rejection

25-2B°C, 47 psig
UMlO UM2 LIMO5

Raw Feed Pemreate Permeate Permeate
PH 9.9 9.7 9.3 8.5
Color, units 3.190 90 25 20
COD. mg/liter 1,450 540 504 485
BOD. mg/liter 266 214 ___ ___
TDS, mg/liter 2,945 2.044 1,676 1,560

Product Recovery -- 61 BB 99
as percent of feed volume
Average Plux (gsfd) -- 21.4 28.4 16.2
Though membranes have been operative in the pulp and paper industry since
1970, growth has been slow. Economic justification has sometimes been margi-
nal; fouling problems often limit membrane life to less than a year.
Dairy Applications
Cheese Whey Protein Recovery. Perhaps the best publicized application for
UF is in cheese whey processing. “Cheese whey” is the supernatant liquid pro-
duced in the cheese making process after precipitation of casein from milk. There
are two types of whey; “sweet” whey (minimum pH of 5.6) results when rennet-
type enzymes are used to coagulate the casein to form Gouda and Cheddar cheeses;
“acid” whey (maximum pH of 5.1) results from acid-induced coagulation of cot-
tage cheese or casein. The typical composition of cheddar cheese whey and cot-
tage cheese whey is shown in Table 3.7. Whey contains half of the milk solids;
i.e., two gallons of milk (17 pounds) yield one pound of cheese and one pound
of whey solids. Most of the lactose winds up in the whey along with 20% of the
protein.
Table 3.7: Composition of Cheddar and Cottage CheeseWhey
From From
Cheddar Cheese Cottage Cheese
Components % %
Total solids 7.31 6.53
Lactose and lactic acid 5.2. 4.39
protein (amino acid ni¶xogen) 0.87 0.86
Ash 0.53 0.61
Fat 0.71 0.67
The lactose is the prime contributor to the high BOD of the whey stream
(35,000 to 55,000 ppm). The 150 billion pounds of cheese whey produced each
year has become a significant pollution problem choking out aquatic life in
Ultrafiltration 233
many streams and lakes and imposing a severe added burden on sewage treat-
ment facilities (1,000 gallons = the waste from 1,800 people).
The valuable component of cheese whey is not the lactose but the whey
proteins, primarily lactalbumin. The amino acid profile of these proteins is su-
perior nutritionally to casein and is equal to or better than whole egg protein.
The heatdenatured form of these proteins has been manufactured for many
years usually by heating the cheese whey to precipitate the proteins. The prod-
uct was tan colored and completely insoluble. With the advent of UF, these
proteins could be recovered, concentrated and demineralized athermally. The
result was a “whey protein concentrate” (WPC) with improved solubility and
other functional properties (emulsification, foamability, water binding, gelation
and cloud stability).
Unfortunately, UF of cheese whey does little to help the pollution problem
since the lactose passes through the membrane. Initially, a two-step process in-
volving RO, after UF, to recover lactose was envisioned (see Figure 3.90). The
recovery of proteins in the first stage (UF) and lactose in the second stage (RO)
is shown in Table 3.8.s6 The first stage can be justified economically with the
value of the high-grade proteins.
INON‘-PROTEINN.
LACTIC ACID I
1 LOW BOD 3_
WATER
Figure 3.90: Flow schematic of two-stage (UF/RO) membrane process for whey
treatment.
Table 3.8: Retention of Proteins, Lactose and COD in Cottage Cheese Whey
by UF and RO
First Stage Second Stage
Feed Stream Permeate Permeate
Total solids 7.8 6.8 1.8
COD, mm 65,600 54,100 800
Protein and amino acid

N2, I 0.6 0.15 0.002
Lactose, % 3.9 3.5 0.05
Lactic acid, $ 0.52 0.52 0.11

Unfortunately, lactose cannot demand a price sufficient to pay for the sec-
ond stage. Current world production of lactose is less than 5% of that in whey
and future demand shows no significant increase.
The growth of UF for this application has been hindered by the difficulty of
using the permeate profitably. This boils down to a profitable utilization of lac-
tose. Figure 3.91 shows the principal uses currently envisioned” for UF per-
meate.
WHEY PERMEATE
Pag feed
t
E
Oryinglstngly or wth other moterialsl
Crystolltzotaon to cattle hck

I
ANIMAL USE
Reactton wth urea
Reactcon to gwe ammomum lactate
Biomass
c I
+Fermentataon +Anoeroblc - Methone for fuel INDUSTRIAL
+Alcohol
t
bother orgomc chemicals
(eg loctlc ocld. ontiblotlcsl I
DIRECT
*Lactose extroctlon HUMAN USE
*Hydrolysis s~gool~e/Glu~~~~g
I+ deionizotton
+ concentrotlon 1
Confectionery Syrup I
Figure 3.91: Principal uses envisioned for UF permeate.
In Switzerland, 92% of the whey is used for pig feed. Reaction with urea to
form lactosyl Urea permits greater quantities of non-protein nitrogen to be used
in ruminant feeds.
The lactose can also be fermented under aerobic conditions to yield a bio-
mass for animal feed with a crude protein content of about 45%. Several com-
mercial plants in France and Austria are producing biomass in this way, but with
marginal economics.
Anerobic fermentation can be used to produce methane to fuel the steam
boilers in the creamery, or to produce alcohol for fuel. One ton of lactose will
produce a half a ton of alcohol. In 1983, the first fully integrated commercial
plant for converting whey to fuel-grade ethanol was constructed in Manteca,
California. It has the capacity to process nearly 500,000 lb of whey per day pro-
ducing 400,000 gal of anhydrous alcohol. In addition to ethanol, the plant was
designed to produce whey protein animal feed in wet cake form and a high-pro-
tein liquid feed supplement for cattle, hogs and poultry.
Whey wine is an experimental product being test marketed. After deprotein-
izing with UF, the lactose is fermented for about a week by special lactose-fer-
menting yeast. The finished product is a pale-yellow, tart, dry wine with a sub-
dued aroma and bouquet.
Ultrafiltration 235
Hydrolysis of lactose to galactose/glucose enhances its solubility (from 22

to 60%) and sweetness (from 15 to 70%). In 1983, a plant utilizing immobilized
lactase to hydrolyze deproteinized lactose (UF) to galactose and glucose was
constructed in Winchester, KY (Nutrisearch-a joint venture between Kroger and
Corning Glass). A similar plant was constructed in Cheshire, England (joint ven-
ture between Corning Glass and the Milk Marketing Board). The resulting syrup
is used as a sweetener and as a growth medium for baker’s yeast.
Figure 3.92 is a process schematic of the world’s largest cheese plant (at
time of startup in 1985) in Corona, California. The plant utilizes many of the
processes described above producing 5 million lb/year of WPC (by UF) (50 to
75% protein) and 2.2 million gal/year of ethanol for gasohol. The lactose per-
meate from the UF unit is concentrated by RO before use as a fermentation me-
dium to produce alcohol.
Lactic Acid Starter Rennet
CHEDDAR
+ CHEESE
80 m Ibs/yr
WATER 17,4Oi) gph
SPRAY WHEY
b DRIER + PROTEIN
CONCENTRATE
5 m Ibs/yr
I
5%
FERMENTATION+ SEPARATOR. DISTILLATION +ALCOHOL
VATS Alcoho 22 m gal/yr
Figure 3.92: Flow schematic of world’s largest cheese plant (Corona, CA).
Perhaps the greatest impediment to the use of UF is membrane fouling with

a decline in flux; it is particularly severe with cheese whey. Usually a cheese
plant cleans its equipment once a day for sanitary purposes with high and low
pH. Early cellulose acetate membranes could not withstand the standard clean-
ing cycles. Today, however, polysulfone and polyvinylidene fluoride membranes
may be cleaned with hot (6O’C) solutions pH 1-12. In some cases, nonionic de-
tergents (0.1%) are used to remove fat deposits and enzyme detergents are used
to hydrolyze protein deposits. Complexing agents such as EDTA and SHMP are
useful in removing inorganic deposits such as insoluble calcium slats. Sanitation
for all but the polyamide membranes is accomplished with solutions of sodium
hypochlorite (with 100 to 200 ppm available chlorine).
Various pretreatments are sometimes utilized to prevent fouling. We have
already seen that pH has a marked affect on flux (see Figure 3.37) by altering
the solubility of the proteins; pH adjustment can also be beneficial in preventing
precipitation on the membrane.
Some of the best work on fouling and pretreatments has been done by Lee
and Merson at the University of California at Davis.58r5g.They examined mem-
brane deposits with a scanning electron microscope in an attempt to develop
preventive measures against membrane fouling. Systemic prefiltration experi-
ments were conducted6’ to study the effects of removing certain constituents
from the whey. The hypothesis was that small amounts of relatively large-
sized protein structures and microorganisms provide anchor points which facili-
tate attachment of the lactoglobulin proteins. Prefiltration designed to remove
these anchor structures did indeed lead to a higher flux with greater stability.
For example, passing untreated whey through Whatman #I filter paper doubled
the permeation rate of a 10,000 MWCO membrane. They also found that cal-
cium sequestration and raising the ionic strength of the whey increased the flux.
It is expected that new surface treatments of the membrane itself will pro-
vide significant improvements in flux stability. UF should be a standard whey
processing step in the future. Even today it is estimated that UF treats 10 to
15% of the total world production.
Milk Concentration. One of the most promising innovations in cheese mak-
ing is to concentrate the casein and proteins in milk with UF prior to coagula-
tion of the casein. Maubois61 is the inventor of this process. Though the UF per-
meate still contains lactose, all of the proteins are incorporated in the cheese re-
sulting in a 5 to 20% reduction in milk usage depending on the type of cheese.62
A two-fold concentration can double the production capacity of each cheese
vat and reduce rennet consumption by up to 80%. The savings in milk usage
alone can often pay for the UF plant in less than one year.
At first, UF was used to make soft cheese of the Camembert type. Next,
semihard cheese like mozzarella was made and more recently, hard cheeses like
Cheddar. In 1981, 105,000 tons of hard cheese was made by this process. Den-
mark has taken the lead in the production of cheese (primarily fetal by UF; in
1981, 92,000 tons (32% of Denmark’s total cheese production) was produced
by UF.
Because some of the minerals are lost with UF, there are some differences
between traditional cheese and that made with UF. In the U.S., this application
has been hindered by concern over whether cheese made with UF meets the es-
tablished standards of identity. In the Spring of 1985, the FDA said feta cheese
made from UF may be acceptable in the U.S.A.
Ultrafiltration 237
There is economic incentive to concentrate the milk on the dairy farm

rather than at the cheese factory. This reduces hauling costs but requires a
greater capital investment in automating a number of small UF units on many
dairy farms. Pioneered in France by Maubois, the first U.S. commercial study
on-the-farm was initiated at a 900 cow farm near Lodi, CA in late 1984. The UF
project is supervised by R. Zall of Cornell University and is co-sponsored by the
California Milk Advisory Board and Dairy Research Inc. Raw milk is heated to
163’F to stabilize micelles in the casein fraction and seal the whey proteins. The
milk is then concentrated by UF to half of the original volume and is shipped to
several cheese makers producing Monterey Jack, Cheddar and mozzarella. The
permeate (containing lactose and soluble salts) is fed to the cows. Preliminary
reports indicate significant savings due to reduced transportation and refrigera-
tion costs plus a dollar credit for livestock feed.
If fouling of UF membranes is a problem with cheese whey, it is an even
greater problem when concentrating milk. The units on the Lodi farm above are
cleaned as follows:
(1) Flush with permeate for 5 minutes

(2) Circulate warm (1 lOoF) water for 20 minutes
(3) Circulate acid for 30 minutes
(4) Circulate warm water for 20 minutes
(5) Circulate chlorinated (200 ppm) alkaline solution for 30 minutes
(6) Rinse with water for 20 minutes
A considerable improvement in UF flux (+lOO%) and stability can be

achieved by prefiltering the milk with a 0.4 /.4pore size MF membrane to remove
bacteria, fat, lipoproteins and coagulated (denatured) proteins. The MF must be
operated in cross-flow to prevent plugging.
There are other markets for concentrated milk besides cheese making. Con-
densed milk or dietary milks (without lactose or salts) can be easily prepared
with UF. The solids concentration can be increased to well over 20% but with
large increases in viscosity (over 30,000 cp).
Food and Beverage Applications (Non-Dairy)

Soy Whey. The pollution problem facing cheese makers is also a problem
for food processors isolating protein from soybeans, cottonseed and other oil-
seeds. Purification of soy protein from defatted soy flake involves extraction and
precipitation. The supernatant after precipitation is similar to cheese whey ex-
cept that the protein concentration is only one-third to one-half of that in cheese
whey. These low protein concentrations reduce the incentive to use UF; as a re-
sult, very few plants use it.
Egg White. The wide spread use of egg white in the baking and candy indus-
tries (more than 300 million pounds in the U.S. per annum) arises from its abil-
ity to form stable foams which can support relatively large quantities of sugar
and/or flour. Fresh egg white contains approximately 88.5% water; it is desir-
able to remove some of this water to save processing, packaging, refrigeration
and transportation costs. However, the proteins are sensitive to heat. Concen-
tration by evaporation or spray drying destroys the functional properties of the
proteins by thermal denaturation. Athermal concentration by UF is ideal.

UF is more appropriate than RO for this application because glucose and in-
organic salts pass through the membrane. Glucose reacts with the amino acids in
the proteins (the “Maillard reaction”) resulting in the development of caramel-
like flavors and odors, evolution of CO* and the eventual separation of an air-
insoluble, dark brown, humus-like material.56
Several alternative methods have been developed to retard this browning re-
action. The most prevalent method is to reduce free glucose by a controlled bac-
teria or yeast fermentation to gluconic acid. The fermentation takes anywhere
from 2 to 72 hours and adds considerable expense. With UF, protein concentra-
tion and glucose reduction can be accomplished simultaneously.
UF concentrates have been shown to have equivalent functional properties
with fresh egg white. ” Even dried solids obtained by spray drying the UF con-
centrate show improved characteristics over that using spray drying alone.63
Economic analysis64 indicates that savings in transportation and drying costs
result in a 40% return on the capital investment required for the UF plant.
Gelatin. Gelatin is a colloidal protein substance whose principal value de-
pends on its coagulative, protective and adhesive powers. It swells up and absorbs
5 to 10 times its weight of water. Water containing only 1% gelatin by weight
will gel when cold. There are threedifferent typesof gelatin: (I) edible, (2) photo-
graphic and (3) inedible (i.e., glue). All types are derived from hydrolysis of col-
lagen (MW 100,000). Since the hydrolysate contains only 3 to 15% gelatin, it
must be concentrated. The traditional method has been steam evaporation fol-
lowed by drum dryers. Again, these elevated temperatures degrade the product
with a resulting loss in gelling power. In addition, the salts in crude gelatin tend
to degrade the polymer structure and are commonly removed by ion exchange.
UF can perform the required concentration at lower temperatures with si-
multaneous removal of salts.65 It is of course true that even with UF elevated
temperatures (above 5O’C) must be used to keep gelatin liquid. It is advanta-
geous to operate at 70°C to take advantage of the lower viscosity and higher flux.
Fruit Juice. Apple juice processors are using UF to increase juice recovery,
improve clarity, reduce materials cost, and to cold sterilize for aseptic packaging.
Before the advent of UF, juice precessors hydrolyzed cloud-stabilizing poly-
saccharides such as pectin and starch with pectinase, coagulated with gelatin, and
then clarified the cloudy cider with a diatomaceous earth filter. The materials
collected on the filter provided an ideal substrate on which to grow and multiply
bacteria. This resulted in high microbial counts in the juice, but high tempera-
ture pasteurization is not desirable because of its effect on flavor.
UF can remove the colloidal haze in apple juice (due to pectins, starch or
protein) while all the sugars (fructose, glucose, sucrose) and flavors (sorbital,
esters, organic acids) freely permeate the membrane. In the process, all bacteria
are removed to produce a clear “sparkling” apple juice which has a long shelf
life.
Table 3.9 is a comparison of operating costs for a 70,000 GPD plant replac-
ing the conventional filter press with UF. The major savings are associated with
the elimination of diatomaceous earth filtration and the reduction in enzyme
(pectinase) requirements. Even with amortization, the operation costs with UF
are only 45% of the conventional system costs. The ROI is 42% with a payback
in 2.4 years.
Ultrafiltration 239
Table 3.9: Operating Costs for Clarification of Apple Juice (Conventional

Filter Press vs UF)
BASIS: 70,000 GPD (20 HR. DAY)
OPERATING COSTS / IMY ($1
CONVENTIONAL
VF FILTER PRESS
-
PECI'INASE 363 w Gelatin 1209
M!mB. REPLACB.(3 YRS) 190 polish filter 88
ROUGH FILTER DE 160
POLISH FILTER OE 195
LABOR 40 480
MAINTENANCE 40 60
POWER 58 88
CLEANING CXiBM. 39 25
730 2305
VEP/FINANCING 518 450
TOTAL (SlS/lOOO GAL)$1248 45% of $2755
CAPITAL 350,000 PAYBACK 2.4 YRS.
INSTALLATION 20,000 ROI 42x

$370,000($5.29/GPD) After Tax
It is possible to eliminate pectinase altogether, but the UF flux decreases

and .a. larger plant is required. Figure 3.9366 shows the effect of “depectiniza-
tion” on the flux, Adding pectinase precipitates soluble pectin into colloidal
particles which are more easily ultrafiltered.
x CENTnIFuoED SOLIOS
Figure 3.93: Effect of pectin on flux in UF of apple juice.

In some cases, the UF concentrate may be diafiltered (see the section on

pharmaceutical applications) to recover the sugar left in the concentrate.
UF produces a superior product with a shelf life of up to 24 months with
“sparkling” clarity. The UF permeate will typically read from 0.4 to 0.6 NTUs
on a turbidimeter, while conventional juice after DE filtration ranges from 1.5
to 3.0 NTUs. Product yields with UF are typically 96 to 99% compared to 80
to 94% without. Cranberry juice and other juices can be clarified in the same
way.
UF of raw sugar juice yields a clear, colloid-free filtrate from which sucrose
can be directly crystallized in high yield and purity.67 In some cases, the UF
permeate is then concentrated by RO; this reduces fouling in the RO module
and permits a higher RO flux.
Wine. MF membranes (0.45 /.r and above) have been used for years to cold
sterilize wine and to remove particles and haze. The recent interest in UF of
wine is surprising; in the past, winemakers have been reluctant to try even a
0.2 /J pore size for fear it would affect the taste. Yet, several wines processed
with UF have recently won awards as some of the best wines in their class.
The current interest in UF is primarily because of its ability to replace sev-
eral processing steps with a single filtration. The haze in wine is often due to
heat-unstable proteins, oxidized/polymerized phenols, gums and pectins. Ben-
tonite and gelatin are added as fining agents to precipitate colloidal proteins so
they can be removed with conventional filters. With UF, precipitation is not
necessary. It can remove all colloids, tannins, carbohydrates and proteins nor-
mally removed by diatomaceous earth and EK pads after fining. In addition,
UF can remove the yeast and/or bacteria normally removed by MF.
Some wineries have tried UF as a last resort when wines show recurring
heat stability problems. Taste panels have been surprised to find that a UF-
processed wine had cleaner nose and better body; the selection of the best
MWCO has been determined by the taste panel. In one case,6s a 10,000 MWCO
membrane was chosen to filter a Sauvignon Blanc after the taste panel noticed
a significant difference in body with a 1,000 MWCO membrane and the lab de-
termined that a 100,000 MWCO membrane did not provide the heat stability
desired.
Press wines are sometimes slightly brown due to higher levels of oxidized
phenolics; UF can remove these color bodies along with the slight bitterness
from phenols. As a rule, UF appears to accentuate acidity and the lighter fruity
elements while eliminating tannic-related “off notes”. UF can also be used to in-
crease the tannin content in red wines with poor body or mouth feel (tannin is
retained by the membrane).
Often, MF is kept as a final filter for insurance. UF prolongs the life of this
filter indefinitely. In spite of DE and pad filtration, MF can clog within 4 to 5
hours without UF.
Beer. There are two primary applications for UF emerging within breweries:
treatment of tank sediment and alcohol reduction.
A number of different kinds of filtration have been used to extract beer
from tank sediment-including yeast presses and DE filtration. The beer recov-
ered with conventional filter techniques is not of good quality. UF produces a
sterile permeate of excellent clarity. Membrane life for tubular systems appears
to be about 9 months.
Ultrafiltration 241
There is an increasing demand for low-alcohol beers which have the beer
taste but with fewer calories and not enough alcohol to cause drunkenness. Beer
fermented to a low alcohol strength does not develop a satisfactory body, taste
and flavor (too little wort). Using UF in the diafiltration mode (see the section
on pharmaceutical applications), the alcohol can be reduced to any desired level
from the same type of beer. In “diafiltration”, as the water-alcohol mixture
passes through the membrane, the permeate volume is replaced by an equal vol-
ume of pure water so as to not change the concentration of the retained com-
ponents in the beer.
Corn Starch. UF can recover colloidal starch and other high molecularweight
compounds that contribute to the high COD level of the waste effluent from a
corn starch plant. Unfortunately, the value of the materials recovered is not
high. The plant must be justified on the basis of pollution abatement.67
Pharmaceutical and Biotechnology Applications

The diversity of applications for UF in the pharmaceutical industry is un-
equaled in any other industry group. Concentration, purification, desalting,
fractionation and sterilization are all practiced in one form or another. In some
cases, the product is in the retentate; in others, it is in the permeate. Occasion-
ally, both retentate and permeate contain valuable products. The fact that all of
these operations are possible with UF at ambient temperatures without phase
change or addition of chemicals/solvents makes it an ideal separation tool for
labile drugs and biologicals. Proteins, polysaccharides, vitamins, hormones, vi-
ruses, vaccines and antibiotics are all processed with UF. Even the water to make
up these pharmaceutical solutions is often sterilized and depyrogenated by UF
(see the discussion on ultrapure water in an earlier section of this chapter).
One of the fastest growing markets for UF and chromatography is in bio-
technology. Broadly defined, biotechnology includes any technique that uses liv-
ing organisms or parts of organisms to make or modify products, to improve
plants or animals, or to develop microorganisms for specific uses. In this sense,
biotechnology has been practiced by man since the dawn of civilization in bak-
ing, brewing, and cheese making. Today’s new biotechnology focuses on applica-
tions for agriculture and industry. To date, most applications in industry are part
of the pharmaceutical industry, which is why we are considering pharmaceutical
and biotechnology applications together.
“New biotechnology” focuses on the industrial use of recombinant DNA
(r DNA) and cell fusion. For example, “old biotechnology” would produce wine
by fermentation but “new technology” seeks to use r DNA to produce wine
with a higher alcohol content. Genetic engineering can now take a gene from one
organism and slip it into the genetic material of another organism. This repro-
grams the microbe to perform differently and creates a hybrid without the time
delays and chance which occurs in cross-breeding of plants and animals by tra-
ditional means. This makes possible the synthesis of new hormones, antibiotics,
insulin and interferon.
The commercial success of r DNA and cell fusion technologies is dependent
on advances in “bio-process engineering”. Most industrial biological syntheses
are carried out on a batch scale, with a small amount of product recovered from
large quantities of water, cellular components, nutrients, and wastes. It is in this
recovery and/or concentration and purification of products where membranes

may be most useful.
The two most important separation techniques for biotechnology appear to
be membranes and chromatography. Membrane separations are more easily scaled
UP and more amenable to continuous processing at less expense than chromatog-
raphy. On the other hand, chromatography can separate proteins and other
macromolecules with greater specificity. Ion exchange chromatography (sepa-
rating on the basis of electrical charge), high performance liquid chromatography
(Using gel exclusion principles depending on the pore sizes of the beads), and
affinity chromatography (reversibly binding some proteins with very specific

ligands/antibodies) can fractionate proteins with a precision unknown in current
UF technology (see the section on fractionation of solutes). However, antibodies
and antigens have been immobilized on both MF and UF membranes and are
sold as immuno affinity membranes for immunodiagnostic tests. These mem-
branes have the potential of combining the specificity of affinity chromatogra-
phy with the low-cost processing of UF.
Cell Harvesting. In Chapter 2, the application of cross-flow MF to cell her_
vesting was discussed. Where the products of the fermentation can pass through
the UF membrane, it is preferred over MF for cell harvesting, The reason is that
asymmetric UF membranes are less prone to fouling and internal pore plugging
than are MF membranes.
Enzyme Concentration and Purification. The number of existing enzymes is
estimated to be more than 10,000 of which more than 100 have been purified in
crystalline form and over 600 in fairly purified form. The molecular weight of
enzymes varies from 12,700 (ribonuclease) to over l,OOO,OOO (L-glutamate de-
hydrogenase, d-carboxylase). All enzymes are proteins, conjugated proteins or
metalloproteins containing one or more active sites per molecule.
Enzymes are highly specific biological catalysts which have an optimum pH
and temperature (usually fairly mild conditions) for maximum activity. From a
chemical engineering point of view they are the ideal catalyst. Inorganic cata-
lysts are seldom as specific (enzyme catalysts yield almost no by-products) and
require high temperatures and/or extremes in PH to be effective.
Because of their protein structure, enzymes may be inactivated temporarily
or permanently by heat, chemicals, or shear forces. UF systems can be designed
to recover enzymes athermally with minimum shear inactivation (using dia-
phragm pumps).
The majority of microbial enzymes are produced in relatively small batch
fermenters. After separation from the biomass with open membranes, filter
presses, or centrifuges, UF is used to concentrate the enzyme and remove small
peptides, oligosaccharides and salts.
Crude extracts of enzymes are also obtained from macerated tissues, germi-
nation seeds, fermented bran, or the “beer” from the submerged fermentation of
selected strains of microorganisms. After extraction and prefiltration to remove
suspended solids and particulates, UF is applied as above.
Blood Plasma Processing. Purified therapeutic proteins are derived from hu-
man blood plasma via cryoprecipitation followed by sequential precipitations
effected through increasing ethanol concentrations at controlled temperature,
PH end ion composition (“Cohn precipitation”). As discussed in the section on
fractionation of solutes, UF cannot be used to fractionate the various plasma
Ultrafiltration 243
proteins; but it is used in subsequent processing steps. Indeed, the concentration

and purification of the various protein fractions was among the earliest applica-
tions for UF in the pharmaceutical industry.
When the ethanolic precipitates are separated from their supernates, they
must be redissolved and concentrated. Further, ethanol and salts must be re-
moved prior to sterilizing filtration with MF and dispensing into containers suit-
able for I.V. delivery. UF is preferred for protein concentration, ethanol removal
and desalting; there is less denaturation and it is more efficient than gel permea-
tion, vacuum freeze-drying, or thin-film evaporation. A major advantage of con-
centration by UF over conventional evaporation or lyophilization is that thesalts
are not concentrated by the process, but freely pass through the membrane.
Thus, the ionic environment for the protein remains constant. Further, since the
solvent does not undergo a phase change, no latent heat is required, and energy
costs are reduced. Finally, since the product need not come into contact with an
air interface, there is less denaturation.
Albumin preparations are routinely concentrated with a 10,000 to 30,000
MWCO membrane so as to contain up to 25% protein; in some cases, concentra-
tion to 40% has been reported.
Gamma-globulin is also routinely concentrated by UF. Buffer solutions are
used to enhance its solubility; gamma-globulin is more prone to precipitation
than albumin.
UF has also been used to concentrate antihemophilic coagulation factor
VIII (AHF) with no measurable loss in activity; it is particularly susceptible to
thermal and shear denaturation. All of the other plasma proteins, including
antithrombin III have been concentrated as well.
Removal of alcohol salt and other low molecular weight substances can be
done most efficiently with UF in the diafiltration mode.
Diafiltration. Diafiltration refers to the process by which UF can rapidly
and efficiently remove salts and/or lower molecular weight species from larger
macromolecules. It is commonly used for fractionation as well as desalting and
is practiced both as a discontinuous and as a continuous process.
Figure 3.94 is an example of the discontinuous process occurring in three
steps: (1) concentration, (2) dilution and (3) recovery.
In the first step, the enzyme is concentrated six-fold from 3 to 18%. (Notice
that even though salt is removed in the permeate, the salt concentration remains
at 3% in retentate and permeate since it is freely permeable to the membrane.)
In the second step, the process stream is diluted by a factor of three to reduce
the salt concentration from 3 to 1%. (As a consequence, the enzyme is now also
diluted by a factor of three; from 18 to 6%). In the third step, the enzyme is re-
concentrated three-fold from 6 to 18%. The net result is the enzyme has been
concentrated six-fold while the salt concentration has been reduced three-fold.
Continuous diafiltration is generally more efficient and preferred. In this
approach, a batch of the solution to be desalted is maintained at constant vol-
ume by adding pure water (dialysate) at the same rate permeate is removed. In
this way, the proteins (or other macromolecules retained by the membrane) re-
main at their initial concentration while the salt concentration decreases contin-
uously. This has been called “constant volume molecular washing” because the
salts are washed out of solution. Continuous diafiltration reduces the processing
time required in the discontinuous process.
ULTRAFILTRATION + DIAFILTRATION
334 I/h deionized 0
CONCENTRATION FACTOR, 6:l CONCENTRATION FACTOR, 3:l
diafiltration water
Z
1T c
3
2.
1,000 I/h feed 167 I/h concentrate 167 I/h final L
Enzyme solution 6% T’S 21% TS 501 I/h feed 7% TS concentrate 19% TS
L 2
+ +
3% enzyme 18% enzyme 6% enzyme 18% enzyme 3
CT
3% low molecular weight 3% low M.W. 1% low M.W. 1% low M.W. J
compounds and salts compounds compounds compounds 2
and salts and salts and salts -l
CD
3-
v v 2
833 I/h permeate 3% TS 334 I/h permeate, 1% TS C
0% enzyme 0% enzyme 9
3% low M.W. compounds 1% low M.W. compounds
and salts and salts
Figure 3.94: Flow schematic for discontinuous diafiltration and concentration of enzymes.
Ultrafiltration 245
When microsolute exchange is desired, the dialysate fluid contains the

microsolute intended to replace the microsolute in the starting solution. For
example, diafiltration can be used to exchange one surfactant for another in a
polymer latex to enhance stabilization. In the pharmaceutical industry, one buf-
fer can be exchanged for another using diafiltration.
Fractionation of two macromolecules (with a large difference in molecular
weight) can be accomplished more completely using diafiltration. The entire
fractionation can be carried out at low concentrations, avoiding problems as-
sociated with higher concentrations like low flux or denaturation.
In the past, the most common method used for microsolute removal has
been batch dialysis. The solution to be dialyzed was placed in seamless regen-
erated-cellulose tubing (“sausage bags”) and suspended in the dialysate allowing
salts to diffuse across the membrane. With diafiltration, the same degree of salt
removal can be accomplished much more rapidly with smaller volumes of dialy-
sate. The pressure driven convective transport of solutes across the membrane is
much faster than concentration driven diffusion (particularly at low salt concen-
trations). In addition, with diafiltration, all solutes (salts and alcohol) are removed
at the same rate independent of the size and diffusivity of the various species;
this makes the process more predictable and controllable.
Assuming the microsolute to be removed is freely permeable to the mem-
brane (R = 0), a simple mass balance reveals that solute removal is exponential:
Co “t
(31) In
( )
c, = v,
where C, = the initial microsolute concentration in the batch
Cf = the final microsolute concentration after volume Vt of
wash solution (dialysate) has passed through the batch
Vt = the volume of wash solution (dialysate) added to the batch,
also equal to the volume of ultrafiltrate removed through
the membrane
V0 = the initial volume of the batch.
When microsolutes are to be exchanged or “washed in”, Equation 31 be-

comes:
where Cd = the concentration of microsolute in the dialysate feed (con-

stant)
C = the concentration of the microsolute washed in with vol-
ume Vt
Since continuous diafiltration involves the processing of a fixed volume (V,), it

is convenient to speak of the number of volume turnovers, VJV, required to
achieve a fixed reduction in microsolute content. For example, Equation 31 pre-

dicts that a 3 volume turnover will effect a 95% reduction in microsolutes and a
5 volume turnover will wash out more than 99% of the salts. Equation 31 is
plotted in Figure 3.95.
When the permeable species is partially rejected by the membrane (as in
fractionation), a higher wash volume turnover will be required to achieve the
same degree of clearance. Figure 3.96 may be used to predict solute clearance
for a specified membrane retention (R).
6
z[/
P
5.
a
/.
02040 60 60 90 95
X OF SOLUTE REMOVED
Figure 3.95: Turnover required to achieve a specified reduction in microsolutes

(R = 0) by diafiltration.
Percentof soluteremoved
Figure 3.96: Turnover required to achieve a specified reduction in microsolutes

(having various rejections) by diafiltration.
Ultrafiltration 247
Often, the end result requires both concentration of macrosolute and re-
duction of microsolute concentration as in Figure 3.94. To optimize the proc-
ess, we must find the macrosolute concentration at which the diafiltration proc-
essing time is minimized. If we diafilter at the starting concentration (3% in Fig-
ure 3.94), we will achieve high flux but the volume of dialysate required for the
desired salt removal will be enormous. On the other hand, if we first concen-
trate the enzyme to its final concentration (18% in Figure 3.94), we will reduce
the volume to be processed, but the flux will decrease. Ng6’ and Cooper” have
derived the equation which specifies the optimum concentration (C”) at which
the diafiltration process time is minimum.
%
(33) c* = 7
where C, is the gel concentration and e is exponential e = 2.718.

For example, if the gel concentration (C,) of the enzyme of Figure 3.94 is
27%, the optimum concentration (C”) at which to carry out diafiltration would
be 10%. Thus, the minimum processing time would start with concentration of
the enzyme from 3 to IO%, diafiltration at 10% (requiring 1.1 volume turnovers
to reduce the salt concentration by three-fold-see Equation 31 and Figure 3.95),
followed by concentration of the enzyme from 10% to the final concentration
(18%). Since diafiltration does not change the volume of the batch, the rota/
concentration time would be constant regardless of whether it is carried out in
one step or two steps separated by diafiltration. Because the processing time re-
quired for diafiltration is a minimum at IO%, the time required for diafiltration
at 3% or 18% would be greater.
Virus Concentration. In large scale virus production, concentration of the
virus or its proteins is frequently necessary to obtain workable volumes for sub-
sequent processing. Often continuous-flow zonal centrifuges are used for this
purpose but with a significant loss in biological activity. Most viruses are larger
than 0.01 /.I and can be safely (closed system) concentrated with a 80,000 MWCO
membrane without loss of activity.
Hormones, extracts and a host of other materials can be concentrated or
purified by UF.
Enzyme Reactors. We turn now to new developments, not fully commer-
cialized, which utilize UF membranes in the processing of pharmaceuticals and
biologicals. The first of these is the enzyme reactor.
Enzymes facilitate many biochemical reactions, serving as a catalyst-i.e.,
the enzyme itself is not consumed in the reaction. In fact, it must be separated
from the products of the reaction. In the past, the enzyme was discarded and
fresh enzyme was charged to the reactor with substrate. Obviously, the eco-
nomics of such a reactor can be improved dramatically if the enzyme can be re-
covered and reused. One way of doing this is to immobilize the enzyme on a
column of glass beads; substrate is fed into one end of the column and products
are withdrawn from the other end. Because the enzyme is affixed to the station-
ary phase, it does not contaminate the product and may be used over and over
again.
Enzymes may also be immobilized on membranes in the same fashion. Three
types of enzyme reactors using UF membranes are introduced here. The equa-
tions which describe their operation may be found with a more thorough de-
scription in Chapter 7.
Mobile Enzyme. Figure 3.97 is a schematic of a continuous enzyme mem-
brane reactor where the enzyme is free and mobile. It is, however, retained by a
UF membrane which is permeable to the products of the reaction. (In this scheme,
the enzyme must be larger in size than the products.) The advantage of this con-
figuration is that there is no steric hindrance to the proper orientation of enzyme
molecules with the substrate molecules. The disadvantage is that the enzyme is
more likely to become inactivated due to the continuous pumping recirculating
the enzyme through the UF unit.”
FILTRATION
Figure 3.97: Flow schematic of a continuous enzyme membrane reactor (mobile

enzyme).
Figure 3.98 shows data of Wang” on the continuous hydrolysis of starch to

glucose by glucoamylase The difference in carbohydrate concentrations in the
retentate and filtrate is small because the product glucose is freely permeable
across the 10,000 MWCO membrane. Only the larger dextrins entering the cell
as substrate are rejected.
IO I I
RETENTATE -
7
l-6 -A &A -
E
FILTRATE
5 (100% GLUCOSE)
7
e6
::
8
a
8% feed concentration, PM-10
$4- membrane at 15 psi, 400~(99)(105) _
g
2 I
0 IO 30 40
TIEl(HOllAS1
Figure 3.98: Continuous hydrolysis of starch by glucoamylase in membrane

reactor.
Ultrafiltration 249
Figure 3.99 shows Wang’s data on the hydrolysis of starch by o-amylase

using the same MWCO membrane. Here the difference in carbohydrate concen-
trations between retentate and permeate is much larger because the products of
a-amylase hydrolysis are predominantly maltose and larger dextrins. Eventually
the dextrins are hydrolyzed to glucose, but at a slower rate. Thus, the carbohy-
drate concentration in the reactor increases with time. These results suggest that
the selection of membrane MWCO can control the products of the reaction.
0.4% feed concentration,

PM-10 membrane at 15 psi,
4OOC(99)(105)
I t 1 I I
20 40 60 SO 100
TIME (HOURS)
Figure 3.99: Continuous hydrolysis of starch by a-amylase in membrane reactor.
Immobilized Enzyme. Figure 3.100 is a schematic of a continuous immobi-

lized enzyme membrane reactor. Here the enzyme is immobilized within the
sponge wall of UF hollow fibers. The membrane skin, on the inner wall of the
fiber, acts as a retentive barrier to enzyme transport into the lumen of the fiber.
This means it is not necessary to immobilize the enzyme onto the membrane sur-
face unless deactivation results. (If immobilization is desirable, glutaraldehyde
may be used to crosslink the enzyme or cyanogen bromide may be used to co-
valently bond the enzyme in place.)
VI PROCESS LINE VI
PROCESS BLEED
(LUMEN ~1021
I”’
SUBSTRATE OR
ENZYME FEED
PROCESS
PUMP
Figure 3.100: Flow schematic of continuous immobilized enzyme membrane

reactor.
This reactor can be operated in two ways. When the substrate is fed to the
shell side of the hollow fiber module (“back flush mode”), the substrate comes
in contact with the enzyme in the fiber wall and product passesinto the lumen
of the fiber from which it exits the module. When the substrate is fed to the lu-
men of the fibers (with all permeate ports closed), it will pass from the lumen to
the shell side where it contacts the enzyme and products will recycle back to the
lumen (“recycle mode”) (see Figure 3.72). The “recycle mode” has the advan-
tage over the “back flush mode” in that the substrate does not have to be free
of suspended matter. In the “back flush mode”, particles in the substrate would
plug the sponge wall.
Breslau and Kilcullenn have compared the “back flush mode” with the “re-
cycle mode” on hydrolysis of lactose with lactase. Table 3.10 shows their re-
sults. Note they operated the “back flush reactor” in both recirculating and sin-
gle pass modes. If there is significant product inhibition, a single pass may be re-
quired. To achieve higher conversions, the product stream may be taken through
valve V 6 to subsequent stages to get 4 passesas in the table. Both modes of op-
eration can achieve the desired conversion.
Table 3.10: Lactose Hydrolysis Via Hollow Fiber Reactors
A. Backflush Reactor
Fiber 0iameter:l 20 nil 45 ml1

Loading Density:* 0.9 gms lactose/ft* 5.45 gms lactose/fte
Substrate: 4.5% Lactose 10.0% Lactose
Method of Operation: Recirculating Single Pass
Time Conversion, Number of Conversion,

Min. % Passes x
0 0.0 0 0.0
20 5.3 1 21.1
60 9.2 2 30.8
90 11.0 3 36.0
1440 40.9 4 40.0
B. Recycle Reactor3
Fiber Diameter:1 45 mil

Loading Density:* 5.45 gmslactose/ft2
Substrate: 10.0% lactose
Method of Operation: Recirculating
Tims Conversion, Time Conversion,
Min. % Min. x
0 0.0 120 17.7
15 3.8 180 22.0
60 10.6
Footnotes
1. Fibers used in this study were 20 ml1 diameter XH50 and 45 nil diameter
XMSO. These fibers have a,molecular weight cutoff of 50,000. Fibers
with molecular weight cutoffs of 10,000 and 5.000 are also available.
2. Wallerstein lactase (200.000 LU/gram activity)
3. The recycle reactor was operating at 25 psi inlet, 23.5 psi outlet
in this study which gave an average driving force for permeate flow
of less than 1 psi.
Ultrafiltration
251
Disperse Soluble Immobilized Enzyme. As we have seen, enzymes immobi-

lized on a membrane have increased stability. The primary disadvantage to im-
mobilization is the restricted access of the enzyme to the substrate. Dunhill et
aI% have explored the immobilization of enzymes to mobile supports. Specifi-
cally, they have attached the enzyme to a soluble high MW polymer which
tended to stabilize the enzyme while reducing steric hindrance and increasing
access of enzyme to substrate. For example, chymotrypsin has been attached
to a soluble dextran using 2-amino-4,Sdichloro-s-triazine as a coupling agent.
This complex was retained in a reactor by a 100,000 MWCO UF membrane
in a configuration similar to that shown in Figure 3.97. Table 3.11 shows the
relative activities as a result of immobilization. Though less than the free en-
zyme, the dextran enzyme complex had more than six times the activity of the
enzyme immobilized on solid DEAE-cellulose.
Table 3.11: Relative Activities of Immobilized Chymotrypsin
Substrate
ATEE Casein
Chymotrypsin 100 100
Dextrhn - Chymotrypsin 81 73
DEAE - Cellulose - Chymotrypsin 13 7
Enzyme stability at 40% is demonstrated in Figure 3.101 for the complex.

A reactor was operated continuously for two weeks at 2O’C hydrolyzing casein
with no loss in enzyme activity.
EFFECTOF INCUBATIONTIME ON ACTIVITY

AT 40'C19
- CEl.L"‘OSE
- CHYEKITRYPSIN
DExTRAN- CHWOTRYPSIN
CHYP~OTRYPSIN
0 10 20 30 40 50 60 70
INCUBATION
TIME (HOURS)
Figure 3.101: Immobilized enzyme stability (chymotrypsin).

For those reactions where the products are larger in molecular weight than
the enzyme, one is forced to go to an immobilized enzyme system or to this
system where the dextran-enzyme complex can be made larger than the prod-
ucts.
Membrane Fermentors. Many of the advantages of enzyme membrane re-
actors are applicable to fermentors. In some cases, a fermentor is simply an en-
zyme reactor using intracellular enzymes rather than extracellular enzymes. En-
zymes are usually more stable within the cell wall and the living cell can regen-
erate itself along with the enzyme. A possible disadvantage is that the enzyme
activity per unit reactor volume may be less in the cells than that of a pure en-
zyme preparation.
There are two primary types of membrane fermentors. These are discussed
more thoroughly in Chapter 7.
Mobile Cells. Figure 3.102’s shows one of the earliest concepts of a con-
tinuous membrane fermentor. The membrane retains the biomass while products
of the fermentation are continuously withdrawn through the membrane. In ad-
dition, metabolic waste products, which inhibit cell growth, are removed con-
tinuously through the membrane, thereby improving cell growth and fermentor
productivity.
Prrrrurircd Air
Fermentation Tank
’ Culturs Conccntrrtion
/ o
l OWl.ml of Producl
- PM.30 Membrane
3 Crude Product. for
M Funher Processing
Figure 3.102: Flow schematic of continuous membrane fermentor.
The increase in cell production due to UF is shown in Figure 3.1O3.76

Clostridium histolyticum is cultivated in batch and continuous fermentors. In
the batch system, the organism approaches a stationary phase at about the
twelfth hour due to toxic metabolite production. In the continuous fermen-
tor, fresh substrate is fed continuously and the toxic products are continu-
ously removed enabling growth to continue well past the fortieth hour. Like-
wise, the enzyme production from this fermentor is enhanced (see Figure
Ultrafiltration 253
3 . 104) .76 In the stationary phase of the batch fermentation, enzyme activity
begins to decrease due to an increasingly unfavorable environment. In the
membrane fermentor, protease (the enzyme) continued to be excreted by the
growing biomass for 50 hours. The ratio of enzyme excreted per unit mass of
cell produced in the membrane system (70 units/mg) was almost twice that
from the batch system (45 units/mg). This would indicate that under the more
favorable environment in the membrane fermentor, the organism is more effi-
cient.
20 I I I I I I
-o--__o
$ 107 _e/--
_ FEED 0
$ - STARTED --‘CONTINUOUS
9 - -0---
5-
h
“J - ” ”
BATCH
if! -
2.
k
)_ HFA-300 membrane at 100 psi,
5 IF 37oc(99)(101)
D
6 0.5 -
I , I I I I
o.20
IO 20 30 40 50 60 70
FERMENTATIONTIME (HOURS)
Figure 3.103: Comparison of batch and continuous fermentors in cell production.
,
‘o--__.
Q-c-
,-o--E&;INUOUS
FEED p”
HFA-300 membrane at 100 psi,

37oc(99)(101)
IO’ I I I I , I J
0 IO 20 30 40 50 60 70
FERMENTATION TIME (HOURS)
Figure 3.104: Comparison of batch and continuous fermentors in enzyme

production.
Other continuous membrane fermentors show similar results. In the con-

tinuous production of ethanol from Zymomonas mobilis, the production rate
was 120 g/hr/Q compared to less than 8 gfhrl!? for the batch system.”
immobilized Cc//s. Cells can be immobilized in the walls of UF hollow fi-
bers and can grow to tissue-like densities (106-10’ cells /cm’) between the fi-
bers. This is about 10 times higher than densities achieved in roller bottles; the
hollow fibers act like natural capillaries in carrying nutrients to the cells and
in removing toxic wastes. Cultures can be maintained for months at a time.
Products like interferon, monoclonal antibodies, antigens, viruses and hor-
mones may be produced continuously with dramatic increases in yield. In ad-
dition, where proteins such as monoclonal antibodies are being produced,
subsequent purification steps can be simplified because the product occurs in
high concentrations with lower concentrations of serum components than is
the case with conventional mouse ascites fluid or suspension culture procedures.
Drioli78r7g has also been successful in casting cellulose acetate and poly-
sulfone UF membranes with a thermophilic microorganism (Caldariella acidd-
phi/a) in the casting solution (30 to 40%). The organism showed good resistance
to organic solvents like acetone used in the casting solution. Drioli reports sig-
nificant activity compared with a similar amount of free cells and no decrease
in activity after 90 days. He has used this bioreactor to hydrolyze lactose to glu-
cose and for malolactic fermentations. For more information, see Chapter 7.
Waste Treatment
Though UF can do little in removing low MW contributors to BOD and
COD, Dorr-Oliver has commercialized a process which couples a UF membrane
to an activated sludge reactor. ” Referring to Figure 3.102, an activated sludge
reactor replaces the fermentation tank. Comminuted raw sewage is fed continu-
ously to this reactor and mixed liquor is continuously circulated through the UF
unit. All of the effluent passesthrough the membrane.
The addition of UF to an activated sludge reactor has a number of advan-
tages.”
(1) Since the membrane retains the biomass, clarifiers are unnecessary.
This also means that the concentration of mixed-liquor suspended
solids in the reactor can be increased from 3,000 to 50,000 ppm,
thereby reducing the size of the reactor. It is also thoughts’ that
the presence of this large biomass absorbs components in the raw
sewage which would normally foul the membrane-perhaps irrevers-
ibly.
(2) Since the membrane is retentive of slowly biodegradable compo-
nents in the sewage, the required detention time may be reduced
or the treatment efficiency at the same contact time may be in-
creased. Essentially infinite detention time is provided for the
slowly biodegradable materials making it possible to metabolize
raw sewage almost completely, with no generation of excess ac-
tivated sludge. Thus, sludge wastage from such a system occurs
infrequently (perhaps bimonthly or quarterly) and in very small
volumes. The system is “self-assimilating”.
Ultrafiltration 255
(3) By virtue of the infinite detention times for large unassimilated

organics and the adsorption of small pollutants within the bio-
mass, the BOD and COD of the effluent is considerably reduced
not to mention the complete elimination of all suspended solids.
DOD born) TSSboml

Conventional
ActivatedSludge <55 <60
Extended
Aeration <30 <70
Bio-membraneProcess <IO 0
The diffusate BOD and TSS values for the biomembrane process
indicate an overall BOD reduction of 95% or better and a 100% re-
duction of TSS.
In addition, coliform bacteria density determinations over a pe-
riod of six months indicated that 90% of the time, the effluent
contained less than 100 coliform bacteria/100 ml.
(4) The addition of powdered carbon to the biomembrane loop fa-
cilitates rapid start up and prevents fouling of the membranes
with oils and greases. Carbon adsorbs soluble organic constituents
and slime generating grease, preventing pollution of the final ef-
fluent (diffusate) until the biomass has developed to the point
where it can digest these pollutants. Furthermore, the activated
carbon acts as a catalyst to enhance the multiplication rate of
the biota population in the system. Thus, start up times are re-
duced from days to hours. Carbon is left in the system to scour
away foulants from the surface of the membrane and is regen-
erated biochemically.
Several bio-membrane plants utilizing this process have been put into oper-
ation by Dorr-Oliver. The oldest were installed in industrial plants in Connecticut
in 1967.
In 1970, a 30,000 gpd plant (15,000 gallons in a 12-hour period) was sup-
plied to the city of Colorado Springs for treatment of sanitary waste originating
at the tourist station located on the top of Pikes Peak. The rapid start up (3 to
4 hours) is ideal for the short tourist season on the mountain. The compactness
of the unit (8’ x 8’ x 20’) is an asset since space is a premium. The reclaimed wa-
ter is not used for drinking, only for washing and flushing toilets, but analysis
shows the effluent is of better quality than the Colorado Springs tap water.
SUMMARY
In its twenty-five year (plus) life, ultrafiltration has been shown to be a re-
liable and economical process for many applications. It has eliminated severe
pollution problems while recovering valuable by-products which have sometimes
paid for the plant in less than a year.
Yet to be solved are the fouling and flux decay problems encountered in the
processing of milk and cheese whey. Permanent surface treatments of mem-
branes by grafting techniques promise to alleviate fouling problems in some in-

stances.
Bioreactors utilizing UF hollow fibers appear to have tremendous potential
for continuous enzyme reactors and fermentors.
As more stringent environmental controls are required and long-term energy
costs rise, there is no doubt that UF will expand rapidly into many new applica-
tions.
REFERENCES
1. Reid, C.E. and Kuppers, J.R., J. Appl. Polym. Sci., 2 (1959) p. 264 ff.
2. Reid, C.E. and Breton, E.J., J. Appl. Polym. Sci., 2 (1959) p. 133 ff.
3. Dobry, A., Bull. de la Societe Chim. de France, 5 (1936) pp. 312-318.
4. Loeb, S. and Sourirajan, S., Sea Water Demineralization by Means of a Semi-
permeable Membrane, UCLA Dept. of Eng., Sea Water Research, Rept.
60-60 (July, 1960).
5. Strathmann, H., Scheible, P. and Backer, R.W., J. Appl. Polym. Sci., Vol 15
(1971) pp. 811-828.
6. Epperson, B.J. and Burnette, L.J., Development and Demonstration of a
Spiral- Wound Thin-Film Composite System for the Economical Produc-
tion of Oxygen-Enriched Air, Final report, Dept. of Eng., No. CSi4029H
(Nov., 1983).
7. Loeb, S., “A Composite Tubular Assembly for Reverse Osmosis Desalina-
tion”, Dept. of Eng., UCLA, unpublished.
8. Grandine, J.D., II, U.S. Patent 4,203,848 (May 20, 1980).
9. Eirich, F.R., So, M.T., Strathmann, H. and Baker, R.W., Polymer Letters,
Vol. 11 (1973) pp. 201-205.
10. Aptel, P., Abidine, N., Ivaldi, F. and Lafaille, J.P., J. Membr. Sci., 22
(1985) pp. 199-215.
11. Shor, A.J., Kraus, K.A., Smith, W.T., Jr. and Johnson, J.S., Jr., J, Phys.
Chem., 72, No. 6 (June, 1968) pp. 2200-2206.
12. Johnson, J.S., Jr., “Polyelectrolytes in Aqueous Solutions-Filtration, Hy-
perfiltration and Dynamic Membranes”, from Reverse Osmosis Mem-
brane Research, ed. by Lonsdale, H.K. and Podall, H.E., Plenum Press,
N.Y. (1972).
13. Leenaars, A.F.M., Keizer, K. and Burggraaf, A.J., Chemtech (Sept., 1986)
pp. 560-564.
14. Porter, M.C. and Nelson, L., “Ultrafiltration in the Chemical, Food Proc-
essing, Pharmaceutical and Medical Industries” from Vol. II of Recent
Developments in Separation Science edited by N.N. Li, CRC Press
(1972) p. 233.
15. Bhattacharyya, D., Schaaf, D.P. and Grieves, R.B., Can. Chem. Eng., Vol.
54 (June, 1976) pp. 185-189.
16. Bhattacharyya, D., Balco, M.G., Cheng, C. and Gentry, S.E., “Use of Neg-
atively-charged Ultrafiltration Membranes”, from Ultrafiltration Mem-
branes and Applications, ed. by A.R. Cooper, Plenum Press, N.Y. (1980)
pp. 605-608.
Ultrafiltration 257
17. Porter, M.C., “Membrane Filtration”, Section 2.1 of Handbook of Separa-

tion Techniques for Chemical Engineers, ed. by P.A. Schweitzer, McGraw-
Hill, N.Y. (1979) pp. 2-3 to 2-103.
18. Breton, E.J., Jr., Water and ion Flow Through Imperfect Osmotic Mem-
branes, Office of Saline Water R&D Progress Report, No. 16, PBI 61391
(1957).
19. Baker, R.W., J. Appl. Polym. Sci., Vol. 13 (1969) pp. 369-376.
20. Wales, M., “Pressure Drop Across Polarization Layers in Ultrafiltration”,
Chapter 11 in Synthetic Membranes, Vol. I, Desalination, A.F. Turback,
editor, ACS Symposium Series 153, (1981) pp. 159-l 70.
21. Tretlin, D.R. and Doshi, M.R., “Pressure-Independent Ultrafiltration-Is it
Gel Limited or Osmotic Pressure Limited”?, Chap. 22 in Synthetic Mem-
branes, Vol. II, Hyper and Ultrafiltration Uses, A.F. Turbak, editor, ACS
Symposium Series 154 (1981) pp. 373-409.
22. Aptel, P. and Clifton, M., “Ultrafiltration”, from Synthetic Membranes:
Science, Engineering and Applications, edited by P.M. Bungay, H.K.
Lonsdale and M.N. de Pinho, NATO ASI Series, D. Reidel (1986) pp.
249-305.
23. Brown, C.E., Tulin, M.P. and Van Dyke, P., Hydronautics, Laurel, Md.
(May, 1970).
24. Porter, M.C., lnd. Eng. Chem. Prod. Res. Dev., Vol. 11 (Sept., 1972) PP.
234-248.
25. Graetz, L., Ann. der Physik und Chemie, Vol. 18 (1883).
26. L&&que, M.A., Ann. Mines, Vol. 13 (April, 1928) pp. 201; 305-308.
27. Dittus, F.W. and Boelter, L.M.K., Univ. Calif. Berkeley, Publ. Eng., Vol. 2
(1930) p. 443.
28. Dravid, A.N., Smith, K.A., Merrill, E.W. and Brian, P.L.T., “The Effect of
Secondary Fluid Motion on Laminar Flow Heat Transfer in Helically
Coiled Tubes”, paper presented at 68th Nat. A.I. Ch. E-. meeting, Hous-
ton, Texas (1971).
29. Sober, H.A., ed.; Handbook of Biochemistry, Chemical Rubber Co., Cleve-
land, OH (1968).
30. de Filippi, R.P., Goldsmith, R.L., “Application and Theory of Membrane
Processes for Biological and Other Macromolecular Solutions”, from
Membrane Science and Technology, ed. by J.E. Flinn, Plenum Press,
NYC (1970) pp. 33-46.
31. Brandup, J., Immergul, E.H.,Po/ymerHandbook, Interscience, NYC (1967).
32. Blatt, W.F., Dravid, A., Michaels, A.S. and Nelsen, L., “Solute Polarization
and Cake Formation in Membrane Ultrafiltration: Causes Consequences,
and Control Techniques,” from Membrane Science and Technology, ed.
by J.E. Flinn, Plenum Press, NYC (1970) pp. 47-97.
33. Eirich, R.R., Rheology, Theory and Applications, Vol. 4, Chapter 2, Aca-
demic Press, NYC (1967).
34. Palmer, A.A.,Amer. J. Phys., 209 (1965).
35. Segrh, G. and Silberberg, A., Nature, 189 (1961) 209.
36. Segre, G. and Silberberg, A., J. Fluid Mech., 14 (1962) 13.
37. Brandt, A. and Bugliarello, G., Trans. Sot. Rheol., 10 (1) (1961) pp. 229-
251.
38. Bixler, H.J. and Rappe, G.D., U.S. Patent 3,541,006 (Nov. 17, 1970).
39. Green, G. and Belfort, G., Desalination, 35 (1980) pp. 129-146.
40. Robertson, G.H., Olieman, J.J. and Farkas, D.F., Concentration Polariza-
tion Reduction in a Centrifugally Driven Membrane Separator, Western
Regional Research Center Report, U.S. Dept. of Agriculture, Berkeley,
CA (August, 1981).
41. Henry, J.D., Lawler, L.F. and Kuo, C.H.A., A./.Ch.E. J., 23, 6 (1977)
pp. 851-859.
42. Parkin, M.F. and Marshall, K.R., N.Z. Jl. Dairy Sci. Technol., 11 (1976)
pp. 107-I 13.
43. Velicangil, 0. and Howell, J.A., Biotechnol. Bioeng., Vol. XIX (1977)
pp. 1890- 1894.
44. Howell, J.A. and Velicangil, O., J. Appl. Polym. Sci., Vol. 27 (1982) pp.
21-32.
45. Kaszyski, M.J., Jones, G.D., Gagnon, S.R., “The Role of Ultrafiltration in
the Production of Ultra-High Purity Water”, paper presented at 41st
Annual Meeting of the International Water Conference of the Engineers
Society of Western Pennsylvania (1981).
46. Center for Disease Control, Informal Quarterly Report of April-June,
7975, U.S. Dept. of Health, Education and Welfare, Public Health
Service (July, 1975) p. 8.
47. Good, C.M. and Lane, H.W., The Biochemistry of Pyrogens, Bul. of PDA,
Vol. 31, No. 3 (May-June, 1977) pp. 116-135.
48. Abramson, D., Butler, L.D. and Chrai, S., J. ParenteralSci. and Tech., Vol.
35, No. 1 (Jan.-Feb., 1981) pp. 3-7.
49. Henderson, L.W. and Beans, E., Kidney International, Vol. 14 (1978) pp.
522-525.
50. Breslau, B.R., Testa, J.A., Milnes, B.A. and Medjonis, G., “Advances in
Hollow Fiber Ultrafiltration Technology” in Ultrafiltration Membranes
and Applications, ed. by A.R. Cooper, Plenum Press, NY (1980) pp.
109-127.
51. Porter, J.J. and Brandon, C.A., Chem. Tech. (June, 1976) pp. 402-407.
52. Cooper, A.R. and Booth, R.G., J. Appl. Polym. Sci.. Vol. 23 (1979) pp.
1373-l 384.
53. Wiley, A.J., Dubey, G.A., Holderby, J.M. and Ammerlaan, A.C.F., J. Water
Pollut. Control Fed., Vol. 42, No. 8, Part 2 (Aug., 1970) pp. 279-289.
54. Bar-Sinai, Y.L. and Wayman, M., Jour. of TAPPI, Vol. 59, No. 3 (March,
1976) pp. 112-l 14.
55. Beder, H. and Gillespie, W.J., Jour. of TAPPl, Vol. 53, No. 5 (May, 1970)
pp. 883-887.
56. Porter, M.C. and Michaels, A.S., Chemtech., Vol. 1 (April, 1971) pp. 248-
254.
57. Coton, S.G., Jour. of Sot. of Dairy Tech., Vol. 33, No. 3 (July, 1980) pp.
89-94.
58. Lee, D.N., Miranda, M.G. and Merson, R.L., “Scanning Electron Microscope
Studies of Membrane Deposits from Whey Ultrafiltration”, J. Food
Tech., Vol. 10 (1975) pp. 139-146.
59. Lee, D.N. and Merson, R.L., J. Dairy Sci., Vol. 58 (Oct., 1975) pp. 1423-
1432.
Ultrafiltration 259
60. Lee, D.N. and Merson, R.L.,J. FoodSci., Vol. 41 (1976) pp. 403-410.
61. Maubois, J.L., Mocquot, G. and Vassal, L.J., U.S. Patent 4,205,090 (May 27,
1980).
62. Maubois, J.L. and Mocquot, G., J. Dairy Sci., Vol. 58, No. 7 (1975) pp.
1001-1007.
63. Sugihara, T.F., Meehan, J.J. and Kline, L., “Application of lnstantizing
Methods of Egg Solids”, paper at 1962 Annual Meeting-inst. of Food
Technologists, Miami Beach, Fla.
64. Tsai, L.S., Ijichi, J. and Harris, M.W., J. Food Prof., Vol. 40, No. 7 (July,
1971) pp. 449-455.
65. Porter, M.C. and Michaels, A.S.,Chem. Tech. (July, 1971) pp. 440-445).
66. Milnes, B.A., “The Application of Ultrafiltration to Apple Juice Proces-
sing”, paper presented at the Apple Juice Workshop, Cornell-NY SAES
(May 3, 1984).
67. Porter, MC. and Michaels, A.S., “Membrane Ultrafiltration-Part 4-Applica-
tion in Processing Vegetal Foods and Beverages”, Chem. Tech., Vol. 1
(Oct., 1971) pp. 633-637.
68. Gnekow, B., “Protein Stabilization of Sauvignon Blanc by Ultrafiltration”,
paper at 34th Annual Meeting of the Amer. Sot. of Enology and Viti-
cultrue (June 22, 1983).
69. Ng, P., Lundblad, J. and Mitra, G., Sep. Sci., 11 (51, (1976) pp. 499-502.
70. Cooper, A.R. and Booth, R.G., Sep. Sci. and Tech., 13 (81, (1978) pp. 735-
744.
71. Porter, M.C., Applications of Membranes to Enzyme Isolation and Purifica-
tion, Biotech & Bioeng. Symp. No. 3, John Wiley & Sons (1972) pp.
115-144.
72. Butterworth, T.A., Wang, D.I.C. and Sinskey, A.J., Biotechnol. Bioeng.,
Vol. XII (1970) pp. 615-631.
73. Breslau, B.R. and Kilcullen, B.M., “Hollow Fiber Enzymatic Reactors An
Engineering Approach,” paper at Third Int. Conf. on Enzyme Engineer-
ing, Portland, Oregon (Aug. 3-8, 1975).
74. O’Neill, S.P., Wykes, J.R., Dunhill, P., Lilly, M.D., Biotechnol. Bioeng.,
Vol. XIII (1971) pp. 319-322.
75. Porter, M.C. and Michaels, A.S., Chem. Tech. (Jan., 1972) pp. 56-61.
76. Wang, D.I.C., Sinskey, J. and Butterworth, T.A., “Enzyme Processing Using
Ultrafiltration Membranes”, from Membrane Science and Technology,
ed. by J.E. Flinn, Plenum Press, NYC (1970) pp. 98-119.
77. Charley, R.C., “Optimization of Process Design for Continuous Ethanol
Production by Zymomonas Mobilis ATCC 29191”, submitted to Bio-
tech. Letters (1982).
78. DeRosa, M., Gambacorta, A., Esposito, E., Drioli, E. and Gaeta, S., Bio-
chimie, 62 (1980) pp. 517-522.
79. Drioli, E., lorio, G., DeRosa, M., Gambacorta, A. and Nicolaus, B., J.
Membr. Sci., 11 (1982) pp. 365-370.
80. Budd, W.E. and Okey, R.W., U.S. Patent 3,472,765 (Oct. 14, 1969).
81. Porter, M.C., “Membrane Ultrafiltration for Pollution Abatement and By-
product Recovery”, Water AlChE Symposium Series, No. 129, Vol. 69
(1972) pp. 100-122.
82. Stavenger, P., Chem. Eng. Prog., Vol. 67, No. 3 (March, 1971) pp. 30-36.
4
Reverse Osmosis
Richard G. Sudak
INTRODUCTION
The first large industrial application of reverse osmosis occurred in 1970

when a 100,000 gallon per day (GPD) system was placed in operation at Texas
Instruments’ electronics manufacturing facility in Dallas, Texas. In this applica-
tion, the reverse osmosis plant was used to pretreat Dallas municipal watdr which
was being converted to ultrapure process water by ion exchange resins. It was re-
ported by the facility engineer that the use of reverse osmosis increased the yield
in the manufacturing operation to such an extent that the resultant savings paid
for the reverse osmosis plant in about two weeks. Now, virtually every electron-
ics plant in the United States uses reverse osmosis as pretreatment in the prepara-
tion of ultrapure water.
I The worldwide total reverse osmosis operating capacity by the end of 1970
was 880,000 GPD. By the end of 1976, reverse osmosis operating capacity had
grown to 167,000,000 GPD’ and, by the middle of 1980, this capacity had more
than doubled to 390,500,OOO GPD.’ Reverse osmosis continued to expand and
it was reported3 that the operating capacity by the end of 1984 was 524,000,OOO
GPD.
Figure 4.1 shows that the rate of growth in reverse osmosis capacity was
slow between 1970 and 1973. From 1974 through 1980, there was an accel-
erated rate of growth during which time the reverse osmosis capacity increased
from 62,500,OOO GPD to 412,500,OOO GPD or an increase of 560% in the 7-
year period. Between 1981 and 1984 (4 years), the total operating capacity in-
creased from 412,500,OOO to 524,000,OOO GPD, an increase of 27%.
Reverse osmosis finds applications predominantly in the following areas:
Industrial - Used to prepare industrial process water or to Proc-

ess wastes.
260
Reverse Osmosis 261
Irrigation - Used to upgrade waters for agricultural purposes.

Military - Used for military purposes.
Municipal - Used to upgrade waters to municipal drinking water
levels.
Power - Used to prepare process water in electric power sta-
tions.
Miscellaneous - Used for various purposes.
ooo-
YEAR
Figure 4.1: Reverse osmosis capacity vs. year.
Table 4.13 below shows the uses of reverse osmosis as described above.
Table 4.1: Reverse Osmosis Product Water Usage
Product Water Use Capacity.MCD* Percent of Total
Industrial 158 31.5
Irrigation 2 0.4
Military 13 2.6
Municipal 191 38.0
Miscellaneous 10 2.0
Power 55 11.0
Yuma DesalLing Plant 73 14.5
Total 502 100.0
* MGD- One million gallons per day

Reverse osmosis finds much of its application in the production of potable

water for municipal water supplies. However, the use of reverse osmosis to pre-
pare industrial process waters and to treat industrial waste is in a close second
place. Agricultural irrigation accounts for only a very small percentage of total
capacity because current sources of agricultural water are much cheaper than
water from reverse osmosis. The military use of the reverse osmosis process was
small as of the end of 1984 but the U.S. military services have discovered reverse
osmosis to be a process ideally suited to providing drinking water in the field.
The process not only is suitable for converting brackish and seawaters into pot-
able supplies but it will also remove toxic and biological agents which result
from germ, chemical and nuclear warfare. Consequently, military use of reverse
osmosis should increase in the future. Power plants have found reverse osmosis
very useful in preparing boiler feedwater. In Yuma, Arizona, a desalting plant
is being constructed by the U.S. Government to improve the quality of the
Colorado River prior to its flowing into Mexico. That plant alone accounts for
14.5% of the worldwide capacity of the reverse osmosis process.
Reverse osmosis is a process that transforms an unusable water supply
into a usable resource. It is capable of renovating a broad spectrum of feed-
waters from municipal water supplies that need polishing for industrial purposes
to seawater that is refined into a potable water supply. Table 4.23 shows the dif-
ferent types of feedwater being processed by reverse osmosis units. Seawater
is considered to have a nominal total dissolved solids (TDS) content of 35,000
mg/!L Wastewater is from industrial or municipal sources and the TDS is vari-
able. Brackish water is defined for the purposes of Table 4.2 only as a water that
may have a TDS from that of municipal water supplies up to 10,000 mg/Q.
Table 4.2: Source of Reverse Osmosis Feedwater
Capacity,MGD Percent of Total
Sea Water 67.9 13.0
Waste Water 26.5 5.0

Brackish Water 82.0
Total 524.0 100.0
As of the end of 1984, the desalination of brackish water accounted for

82% of capacity. This is due to the fact that early reverse osmosis membranes
were incapable of single stage seawater desalination and, thus, they were limited
to brackish water desalination. Within the last 10 years, significant advances
have been made in both the flux and rejection capability of membranes and re-
verse osmosis is technically able to desalt seawater in a single stage. In the re-
cent past, it has been an effective competitor to the distillation process in sea-
water desalination. In fact, reverse osmosis is now beginning to replace existing
distillation capacity in the Middle East.4
Although reverse osmosis is a relatively new technology, there is sufficient
operating capacity in a number of varied applications to warrant confidence
in the process. From a technical and economic point of view the process is ca-
Reverse Osmosis 263
pable of desalting a broad range of feedwaters from municipal water supplies to

seawater. It has economic viability in a large number of industrial applications.
BASIC PROCESS CONSIDERATIONS
An industrial reverse osmosis plant usually will consist of three separate sec-
tions which are shown in Figure 4.2. The first section is the pretreatment sec-
tion in which the feedwater is treated to meet the requirements of reverse os-
mosis element manufacturers and the dictates of good engineering practice.
Following pretreatment, the feedwater is introduced into the reverse osmosis
section where the feedwater is pressurized and routed to the reverse osmosis
elements which are in pressure vessels.The feedwater flows across the membrane
surface where product water permeates through the membrane and a predeter-
mined amount remains behind as reject. The reject is discharged to waste while
the product water is routed to the posttreatment section. The third or post-
treatment section treats the product water to remove carbon dioxide and adds
chemicals as required for industrial use of the product water.
I I
PRETREATMENT
PRODUCT
* TO INDUSTRY
Figure 4.2: Industrial reverse osmosis.
Although operating reverse osmosis plants are commonplace, the subject

of salt and water transport is still controversial. Several transport models have
been proposed and these will be discussed briefly below. A more thorough
presentation is made in a paper by Soltanieh and Gill.’
The sieve mechanism is the simplest and easiest model to understand. This
model proposes that salt and water are separated due to physical size differences
by a membrane with a pore size that lies between the sizes of salt and water.
While most laymen prefer this concept, it is unfortunate that for solutions, such
as sodium chloride in water, the sizes of the salt and water molecules are almost
the same. This fact would seem to rule out the sieve mechanism model.
Another model proposed is the wetted surface mechanism or the water
clustering concept of transport. It is generally recognized that reverse osmosis
membrane material is wettable and that water tends to be absorbed on the
membrane by hydrogen bonding. In this concept, it is theorized that the water
film at the surface of the membrane obstructs the pores and prevents salt from
entering the pores. The water passesthrough the membrane by passing from one
absorbed site to the next until it reaches the other side of the membrane barrier
layer. The energy requirements for solvent migration are much less than the
energy requirements for salt migration and, thus, the separation of salt from
water takes place.
Another concept of water and salt transport in reverse osmosis is the pref-
erential sorption-capillary flow mechanism. In this model, the surface of a mem-
brane is microporous and heterogeneous at all levels of solute separation. It is
hypothesized that, due to the chemical nature of the membrane skin layer in
contact with the aqueous solution, a preferential sorption for the water causes
a sorbed water layer to be formed at the skin. This layer of purified water is
then forced through the capillary pores by pressure.
The solution-diffusion model of transport assumes a nonporous, homogene-
ous membrane surface layer. Each component in a pressurized solution dissolves
in the membrane and then diffuses through the membrane. The flow of water
and salt through the membrane is uncoupled, i.e., they are independent of each
other, and the water transports through the membrane at a more rapid rate than
the salt.
The product water flow through the membrane is defined as follows:
F, = A*(AP-An)
where F, = water flux through the membrane

A = water transport coefficient
AP = pressure differential across the membrane
An = osmotic pressure differential across the membrane
The flow of water through a reverse osmosis membrane is primarily dependent

on the applied pressure differential and the osmotic pressure differential across
the membrane. The osmotic pressure is directly dependent on the salt concen-
tration of the process stream. ks a rule of thumb, each 100 mg/Q of dissolved
solids is roughly equivalent to one psi of osmotic pressure. Since the product
stream usually has a very low salt content, the osmotic pressure of that stream is
negligible. In addition, the product stream normally leaves the reverse osmosis
pressure vessels at near atmospheric pressure so that the applied pressure differ-
ential is the feed pressure. Consequently, the term “net applied pressure” has
come to mean the applied pressure minus the feed osmotic pressure.
Reverse Osmosis 265
The flow of salt or dissolved solids across the membrane is dependent on

the following equation:
F, = B*AC
where F, = salt flux

B = salt transport coefficient
AC = salt concentration gradient across the membrane
The salt transport is primarily dependent upon the concentration of dissolved

solids on each side of the membrane.
The solution-diffusion model seems to represent the performance of a re-
verse osmosis membrane. Figure 4.3 shows the salt rejection and flux of a low
pressure polyamide membrane as a function of applied pressure. The membrane
was operated on a 5,000 mg/ll aqueous solution of sodium chloride at 25°C. As
can be seen, there was no product water flow until the applied pressure exceeded
the osmotic pressure (50 psi). After this, the flux increased linearly as would
be predicted by the above water flux equation. Rejection is poor at lower pres-
sures and increases rapidly until it reaches an asymptote at an applied pressure
of about 150 psig. This can be attributed to a near constant flow of salt with
a rapidly increasing product water flow which results in a more dilute product
or in increased rejection. These data tend to substantiate the assertion of the
solution-diffusion model that flow is uncoupled.
APPLIED PRESSURE, PSIG
Figure 4.3: Membrane flux and rejection vs. applied pressure.
It is noted that the data shown in Figure 4.3 was derived in membrane test
cells with near perfect small areas of membrane. In a practical system, there
would be a number of imperfections in the membrane and the salt flow through
these capillaries would contribute to the total salt flow. Therefore, a practical
salt transport model must also take into account the contribution of the mem-
brane imperfections to salt flow.
The water transport coefficient (A) is not a constant in that it varies with
temperature. The product flow as a function of temperature may be estimated’s
by using the following equation:
Q 25/ Qt = ex
where Q 25 = flow at 25°C or 77°F

Qt = flow at temperature T, “C
e = 2.71828
X = U [l/(T+273) -l/2981
T = Temperature in “C
u = 2723 (for cellulose acetate membranes).
Figure 4.4 graphically shows the ratio of water flux at 25°C to water flux at
other temperatures.
2.4
2.2
2.0
1.2
1.2
1.4
1.2
1.0
0.2
0.6
0.4
0.2
0 10 20 20 40 60
TEMPERATURE OC
Figure 4.4: Temperature correction factor.
As a rule of thumb, the product water flow with constant net applied pres-
sure will increase about 3% for each degree centigrade increase in feedwater tem-
perature. Salt flux through the membrane is also directly proportional to tem-
perature and the ratio of salt flux to water flux is essentially constant at differ-
ent temperatures. This results in little or no change in rejection as a function of
Reverse Osmosis 267
temperature. For some of the newer composite membranes, the water and salt
permeation coefficients also vary as a function of pH.
Reverse osmosis is a cross-flow membrane separation process which sepa-
rates a feed stream into a product stream and a reject stream. The recovery of a
reverse osmosis plant is defined as a percentage of feedwater that is recovered as
product water. As all of the feedwater must be pretreated and pressurized, it is
economically prudent to maximize the recovery in order to minimize power con-
sumption and the size of the pretreatment equipment. Since most of the salts
remain in the reject stream, the concentration of salts increases in that stream
with increased recovery. For instance, at 50% recovery, the salt concentration
in the reject is about double that of the feed and at 90% recovery, the salt con-
centration in the reject is nearly 10 times that of the feed. In cases of sparingly
soluble salts, such as calcium sulfate, the solubility limits may be exceeded at a
high recovery. This could result in precipitation of the salt on the membrane sur-
face resulting in decreased flux and/or increased salt passage. In addition, an in-
crease in recovery will increase the average salt concentration in the feed/reject
stream and this produces a product water with increased salt content. Conse-
quently, the recovery of a reverse osmosis plant is established after careful con-
sideration of the desired product quality, the solubility limits of the feed con-
stituents, feedwater availability and reject disposal requirements.
While the above equations are helpful in describing the reverse osmosis
process, the water and salt transport coefficients seldom are used to describe
membrane performance. Most manufacturers test reverse osmosis membrane
with standard solutions as described below.
(I) Brackish Water Membrane-The flux and rejection of the mem-

brane is determined when the membrane is tested on a feedwater
of an aqueous sodium chloride solution with a concentration of
2,000 mglQ at an applied pressure of 420 psig (net applied pres-
sure = 400 psig) with a temperature of 25°C and a feed pH of from
5.0 to 6.0 for 30 minutes prior to data collection.
(2) Seawarer Membrane-The seawater test is similar to the brackish
water test except that the feedwater is an aqueous sodium chloride
solution with a concentration of 35,000 mg/R and the test pressure
is 800 psig.
The results are reported as gallons per day per square foot of membrane
area (membrane flux) and as rejection of sodium chloride. Rejection is calcu-
lated as follows:
R = (I-CPKF) 100
where R = Rejection, percent

CP = Product water concentration
CF = Feedwater concentration
The sodium chloride rejection differs from that of other inorganic and or-
ganic dissolved solids, and membrane manufacturers will provide information
and rejection data that are available for their specific membrane. Table 4.3
shows typical results for a composite membrane when tested on a multicom-

ponent solution. The rejection of the divalent ions such as calcium and sulfate
is much better than the rejection of the monovalent ions such as sodium and
chloride. If salt passage is defined as product concentration divided by the feed
concentration, or one minus rejection, then it can be seen that the salt passage
for the divalent ions is about one-fifth of the salt passage for the monovalent
ions.
Table 4.3: ionic Rejections
-IOn Feed, mp,/l Product, mg/l Rejection, X
Calcium 61 0.2 99.6
Sodium 150 3.0 98.0
Potassium 12 0.3 97.4
Bicarbonate 19 0.7 96.2
Sulfate 189 0.4 99.8
Chloride 162 2.9 98.2
Nitrate 97 3.5 96.4
TDS 693 11.0 98.4
The abovedescribed tests are conducted at a low recovery rate to minimize

the effects of concentration polarization which is described below, For example,
membrane tests above are conducted at less than 1% recovery and tests with
spiral wound elements are conducted at recoveries from 5 to 10%.
Reverse osmosis is a cross-flow process and, as in any dynamic hydraulic
process, the fluid adjacent to the membrane moves slower than the main stream.
While the main stream flow may be turbulent, the layer next to the membrane
surface is laminar. This thin, laminar flow film is called the boundary layer.
When water permeates through the membrane, nearly all of the salt remains be-
hind in the boundary layer next to the membrane. The salt must then diffuse
across the boundary layer and back into the bulk stream. This results in a bound-
ary layer with a salt concentration which is more concentrated than the bulk
stream. The effect has been termed concentration polarization, and it is de-
fined by the following equation:
p = CBICM
where /3 = Concentration polarization

CM = Concentration in the main stream
C8 = Concentration in the boundary layer
Concentration polarization increases the salt concentration at the membrane sur-

face, and this results in an increased osmotic pressure at that surface. The in-
creased osmotic pressure causes a drop in water flow as shown in the following
equation:
Reverse Osmosis 269
Fw = A*(P-/3n)
where Fw = the water flow

A = the water transport coefficient
P = applied pressure
P = concentration polarization
n = osmotic pressure of main stream
The increased salt concentration at the membrane surface will also increase the
tendency of sparingly soluble salts to precipitate on the membrane.
The flow of salt also increases and this can be simply shown in the following
equation:
Fs = B*@Cl-C2)
where Fs = salt flux

B = salt transport coefficient
P = concentration polarization
Cl = main stream salt concentration
c2 = product water salt concentration
In the imperfect membrane with a small number of pores, the increased salt con-
centration at the membrane surface would also result in increased salt passage
through the pores which would be directly proportional to concentration polari-
zation.
The information shown in Table 4.4 below assumes a membrane operating
on various feedwaters with a total dissolved solids (TDS) content of 2,000,
5,000 and 35,000 mg/Q. It is assumed that the membrane will deliver 20 gallons
per square foot per day at a 400 psi “net applied pressure” and have a rejection
of 99% when there is no concentration polarization, i.e., fl= 1.0.
Table 4.4: Effects of Concentration Polarization
Feed TDS, mg/l ?,ooo 5,ooo 35,000
Flux, gfd
B = 1.0 20.0 20.0 20.0
8 - 1.1 19.9 19.8 18.3
8 = 1.5 19.5 18.8 11.3
8 = 2.0 19.0 17.5 2.5
Rejection, %
S= 1.0 99.0 99.0 99.0
f3 = 1.1 98.9 98.9 90.0
B = 1.5 98.5 98.4 97.3
8 = 2.0 97.9 97.7 84.0

The penalty of a high concentration polarization is not as severe with a low

TDS feedwater as it is with a high TDS feedwater. The recommendations as to
minimum flows or maximum recoveries which are specified by the reverse os-
mosis element manufacturer should be followed at all times, especially when the
application is the desalination of high TDS waters.
Concentration polarization cannot be eliminated, but it can be minimized
by decreasing boundary layer thickness. This is done by increasing the flow rate
across the membrane surface or introducing turbulence promoters into the
feed/reject stream. In order to achieve optimum performance, most membrane
manufacturers will recommend a minimum feed rate to or from their elements
and a maximum recovery in order to minimize the effects of concentration po-
larization.
REVERSE OSMOSIS MEMBRANES
Osmotic phenomena have been observed since the middle of the eighteenth
century. The first experiments were conducted with animal membranes and it
wasn’t unit1 1867 that artificial membranes were employed. In the early 1950’s,
research workers at the University of Florida demonstrated, with thick films,
that cellulose acetate possessedunique salt and water transport properties which
made it potentially attractive as a reverse osmosis desalination membrane. Dur-
ing the 1960’s, Loeb and others at the University of California at Los Angeles
developed techniques to prepare cellulose acetate membranes with an econom-
ical water flux and salt rejection at moderate driving pressures. With this devel-
opment, reverse osmosis became a practical possibility.
The early cellulose acetate membranes were only suitable for brackish water
desalination as they were not capable of operating at the high pressures required
for seawater desalination (see below). In addition, the membrane rejection was
insufficient to desalt seawater in one stage. The first membranes were fabricated
from cellulose diacetate which is subject to hydrolysis at either a low or high pH,
biological attack and compaction at temperatures in excess of 85°F. Cellulose
acetate also compacts (densifies) at pressures above 400 psig and temperatures
below 85°F. Suitable feedwater pretreatment lessens the adverse effects of tem-
perature, pH and biological organisms to the point where the membrane can be
used in practical reverse osmosis plants.
A considerable amount of research has been done to develop a membrane of
cellulose triacetate as this material is more stable to extremes of temperature and
pH and it will better withstand chemical and biological attack. While the cellu-
lose triacetate membranes operated quite well on a short term basis, they were
prone to compact at an operating pressure of 400 psig, with the resultant loss of
flux to an impractical level, in a short period of time.
It has been demonstrated that a blend of cellulose diacetate and cellulose
triacetate provides an improved membrane in that:
(1) it is more stable than the cellulose diacetate membranes;

(2) it has a better flux and rejection than the cellulose diacetate; and
(3) it is more resistant to compaction than either the diacetate or
triacetate.
Reverse Osmosis 271
Incidentally, all of the cellulose acetate membranes will tolerate a limited

amount of residual chlorine which allows chlorine to be used for feedwater
disinfection.
The cellulose acetate membranes are asymmetric and fabricated from a
single polymer. The use of electron microscopy in the 1960’s demonstrated
that the cellulose acetate membranes consisted of a relatively thin dense layer
and a thicker porous layer of the same material. The membrane thickness is
usually about 100 micrometers with the dense layer accounting for about
0.2% of the thickness and the remainder being an open cell porous matrix (see
Figure 4.5).
/
Porous Matrlx
Figure 4.5: Single component asymmetric membrane.
During the 1960’s, the DuPont Company screened numerous polymers to

determine the suitability of materials other than cellulose acetate for use in re-
verse osmosis desalination. The results of this work indicated that aromatic
polyamides were the “choice as the best polymer type for use in the DuPont
commercial permeators”.’ The company was most successful in developing an
asymmetric aromatic polyamide reverse osmosis membrane in a hollow fine
fiber configuration which successfully competed with cellulose acetate in the
market place.
Shortly after the concept of an asymmetric membrane was established,
composite membrane research was initiated. A composite membrane is also
asymmetric but it consists of two polymer layers which are the membrane
barrier layer and the porous support layer (Figure 4.6). The porous support
is formed separately, by conventional membrane casting techniques, from one
polymer. The porous support has a thickness of between 75 and 100 micro-
meters and its porosity is due to numerous small perforations through the sup-
port. The membrane barrier layer is a dense thin film of another polymer that is
formed or deposited in a subsequent operation on the porous support. The
membrane barrier layer varies in thickness from 400 to 1,000 angstroms.
Membrane Serrler Layer
Polyamide
I 1400-.l.OOOAn~strome
i
s
4”r
P
5
c”
Porous Support
I- 76 Mlcrone
Figure 4.6: Two component composite membrane.
Several polymers have been used as porous supports. One of the earliest
composite membrane systems was a porous support formed from cellulose
nitrate-cellulose acetate with a membrane barrier layer of cellulose triacetate.
While this membrane successfully desalted seawater, it was fragile and expen-
sive. To a large degree, present day commercially available composite membranes
use a polysulfone porous support.
Membrane barrier layers have been formed on porous supports in the fol-
lowing manner:’
(1) solution,
(2) thin film polymerization and
(3) interfacial polycondensation.
The solution coating technique was used in the preparation of the cellulose
triacetate membrane discussed above. A solution of cellulose triacetate in chlo-
roform was deposited on the porous support and the solvent was then evapo-
rated leaving a thin film on the porous support. Thin film polymerization was
used to prepare a polyfuran membrane barrier layer on polysulfone. In this case,
the monomer furfuryl alcohol is polymerized in situ by adjustment of pH and
temperature. This membrane proved to be highly susceptible to oxidizing agents
and is of limited value. By far the most valuable technique in the formation of
membrane barrier layers is interfacial polycondensation. In this method, a poly-
mer is formed on the porous support surface at the interface of organic and
aqueous phases by reaction of specific molecules dissolved in each phase. It is
by this method that a number of polyamides and polyurea membrane barrier
layers have been formed on polysulfone. Elements containing these membranes
are available commercially.
There has been very little progress in the last five years in improving the Per-
formance of single polymer asymmetric membranes. Meanwhile, the composite
membranes have been improved and they exhibit a higher flux and better re-
Reverse Osmosis 273
jection at lower operating pressures than is available with the single component
asymmetric membranes. The polyamide and polyurea composite membranes
can withstand higher temperatures and larger pH variations than the cellulose
acetate based membranes and they are immune to biological attack. On the
other hand, as noted above, the cellulose acetate membranes can tolerate a lim-
ited concentration of residual chlorine while the polyamide and polyurea mem-
brane barrier layers are subject to disintegration by residual chlorine and other
oxidizing agents. Consequently, alternate methods of water disinfection or chlo-
rination followed by dechlorination are employed in reverse osmosis systems
using these membranes.
While the cellulose acetate membranes are compacted at the moderate
pressures required for brackish water desalination, the thin film composite
membranes exhibit ho compaction at these pressures and a very low compaction
rate at pressuresof 1,000 psig.
The superior flux and rejection capabilities of the thin film composite mem-
brane has been demonstrated at the municipal wastewater reclamation facility
of the Orange County Water District in California. Both asymmetric cellulose
acetate and thin film composite membranes were tested on lime clarified second-
ary effluent. The pilot plants were operated at 85% recovery and the rejections
reported in Table 4.5 are the percent rejection of the constituents in the feed-
water and not the rejection of the average concentration of the specific constitu-
ents in the feed/reject stream. Use of the average concentration would give a
higher rejection in both cases.
Table 4.5: Rejection Comparison
Constituent CA Membranetg TFC PolyamideMembrane(lO1
Na 93.8 99.0
NH4 89.0 94.8
SO4 99.3 99.8
Cl 93.3 95.2
NOa 71.0 97.7
COD 93.6 >94.8
TDS 94.4 97.5
The thin film composite membrane exhibited superior overall rejection per-
formance in these tests, with ammonia and nitrate rejection showing an out-
standing improvement. It has also been reported that silica rejection by the thin
film composite membranes is superior to that of cellulose acetate. While the
above data indicates a marginal improvement in the rejection of chemical oxygen
demand (COD), which is an indication of organic content, other tests conducted
by membrane manufacturers show that the polyurea and polyamide membrane
barrier layers exhibit an organic rejection that is clearly superior to that of cellu-
lose acetate. Reverse osmosis element manufacturers should be contacted for
rejection data on specific organic compounds.
MEMBRANE PACKAGING
Reverse osmosis membrane is produced in sheet form-up to 60 inches wide

and lengths up to 1,500 feet-and as a hollow fine fiber. The asymmetric cellu-
lose acetate was originally produced as a sheet and later as a hollow fine fiber.
The asymmetric aromatic polyamide was originally produced as a hollow fine
fiber and later in sheet form. The composite membranes with polyamide or poly-
urea membrane barrier layers are produced in sheet form as of the end of 1987,
but research has been and will continue to be done to produce the composite
reverse osmosis membranes as a hollow fine fiber.
The Office of Saline Water, U.S. Department of Interior, sponsored much of
the development of membrane packaging configurations for reverse osmosis
membrane. The configurations which have been developed and evaluated are as
follows:
(I) plate and frame,

(2) spiralwound,
(3) tubular and
(4) hollow fine fiber.
Plate and Frame

The plate and frame configuration is much like the conventional plate and
frame filtration concept except that, in reverse osmosis, a typically higher fluid
operating pressure is used. Figure 4.7 is a schematic of the plate and frame ap-
proach. The membrane package is installed in a pressure vessel designed and fab-
ricated to withstand operating pressures from 400 to 1,000 psig. A stack of par-
allel porous plates are used to support the membrane on each side of the porous
plate, Feedwater under pressure enters the top of the pressure vessel and flows
between the parallel stacks of membrane/porous plates. Product water passes
through the membrane and into the porous plate to be routed to the product
water collection system and then out of the pressure vessel. As the feed/reject
stream passes across the membrane, it becomes concentrated and eventually
leaves the pressure vessel as concentrate or reject. More advanced designs utilize
external tie bolts which hold together sufficiently thick end plates which contain
the pressurized feed/reject stream. The plate and frame packaging configuration
has an advantage in that only the membrane must be replaced when a membrane
becomes defective. It has disadvantages of complex flow patterns and high costs.
The plate and frame configuration can achieve a packaging density of up to 150
square feet of active membrane area per cubic foot of pressure vessel. Overall,
the plate and frame concept has not been economically competitive and there
are no remaining manufacturers of this type of equipment in the United States.
Spiral Wound
The spiral wound membrane packaging configuration is shown in Figure
4.8. Basically, the spiral wound element consists of two sheets of membrane
separated by a grooved, polymer reinforced fabric material. This fabric both
supports the membrane against the operating pressure and provides a flow path
Reverse Osmosis 275
for egress of the product water. The membrane envelope is sealed with an ad-
hesive on three sides to prevent contamination of the product water. The fourth
side is attached to a product water tube which has perforations within the edge
seal so that product water can be removed from the porous product water car-
rier material. The membrane envelope is rolled up around central product water
tube, with a plastic mesh spacer between the facing membrane surfaces, in a
spiral. The mesh spacer not only serves to separate membrane surfaces but it
provides a flow path for and turbulence in the feed/reject stream of each ele-
ment. The elements have an outer wrap to contain the feed/reject stream in the
mesh passageway and brine seal to insure that the feed/reject stream goes through
the element and not around it. Spiral wound elements are available in lengths
from 12 to 60 inches and diameters from 2 to 12 inches. Packaging densities of
up to 300 square feet of membrane to 1 cubic foot of pressure vessel have been
attained. Reverse osmosis plants with a capacity from 100,000 to 5,000,OOO gpd
of product would normally use elements with an 8-inch diameter by a 40-inch
or 60-inch length.
WI WdlW In
a
Fresh
’ Porous
PIales
Figure 4.7: Plate and frame packaging.

Fcedwalec6rine Spacer Membrane
Feedwaler
to Fresh Water by Passage

through Membrane
Product Waler Flow

(atler Passage Product Tube
through Membrane)
End view
Membranes
PIO~UC~Water Side
Sackinq with Membran
oi Each ‘&da
Figure 4.8: Spiral wound packaging.

Reverse Osmosis 277
Spiral wound elements are installed in a pressure vessel which is usually fab-
ricated from fiberglass reinforced plastic. The pressure vessel inside diameter is
sized to match the outside diameter of the element brine seal. The pressure ves-
sels are designed and fabricated to accommodate from 1 to 6 elements and op-
erating pressures from 50 to 1,000 psig. Figure 4.9 shows a pressure vessel with
6 elements installed. Feedwater enters one end of the pressure vessel and flows
through the first element in which about 10% of the feed permeates through
the membrane and flows through the product water carrier material into the
product water tube. The reject from the first element flows to and through the
second element and the reject from this element becomes the feed to the next
element, etc. The reject from the last element is routed from the pressure vessel
to the high pressure reject manifold. The first and sixth element product water
tubes are sealed to the pressure vessel end caps by O-ring devices to prevent con-
tamination by the high pressure feed or reject to the purified product water
stream. The element product water tubes in the pressure vessel are connected to
each other with interconnectors which again are O-ring devices whose seals pre-
vent product water contamination. The product water can exit the pressure
vessel, usually at near ambient pressures, from either end of the pressure vessel.
In a single pressure vessel with six elements, between 40 and 60% of the feed-
water to the pressure vessel is recovered as product water from a brackish water
feed and from 25 to 35% is recovered from a seawater feed.
The advantages of the spiral wound elements are the high packing density
and high flux which makes it one of the most cost effective elements. The dis-
advantage of the element is that a moderate amount of pretreatment is required
for some feedwaters to prevent fouling of the mesh brine spacers.
Tubular
The tubular reverse osmosis device is shown in Figure 4.10. The tube serves
as the pressure vessel and the membrane is installed inside the tube. Tubes with
inside diameters of % and 1 inch have been used. Uniformly porous fiberglass
reinforced plastic tubes have been used and nonporous but perforated copper,
stainless steel and fiberglass tubes have also been successfully used. The mem-
brane can be bonded to the tube in which case it is cast in situ or the membrane
can be loose. The loose membrane is cast in sheet form and a cylindrical section
is formed and placed in the tube. Packing densities for the %-inch diameter tube
are about 100 square feet per cubic foot and about 50 square feet per cubic foot
for the l-inch diameter tube.
Pressurized feedwater enters the tube through an end fitting which seals the
membrane to the tube and prevents cross contamination of the product water.
The feed water flows down the length of the tube and product water permeates
through the membrane and weeps through the tubular pressure vessel into a col-
lection basin. The reject flows through an end fitting and is routed to additional
tubes in series or to waste.
The major advantages of the tubular reverse osmosis configuration are the
ability to tolerate high suspended solids concentrations in the feed and the pos-
sibility of mechanical membrane cleaning. The disadvantages are the excessive
number of tube end fittings in proportion to the active membrane area in each
pressure vessel, the bulkiness of the reverse osmosis plant and the high cost.
I
End Cap Retaining Ring O-Ring Seal Interconnector 5
g
8
k
0
FRP Pressure Vessel Brine Seal
Figure 4.9: Spiral wound element pressure vessel assembly.

Reverse Osmosis 279
Practical Unit
Salt
Water
In
Fresh Water
Figure 4.10: Tubular packaging.
There have been a number of attempts to commercialize tubular reverse osmosis

systems in industrial applications. As of the end of 1987, there were no large
scale tubular reverse osmosis manufacturers in the United States, although there
are some in Europe and Japan.
Hollow Fine Fiber

Hollow fine fiber synthetic filaments have been prepared for use in the tex-
tile industry for a long time. This well-known technology was adapted to prepare
polymers, which are suitable for reverse osmosis desalination, into hollow fine
fibers. The most readily available polymers in hollow fine fiber elements are
aromatic polyamide, cellulose diacetate and cellulose triacetate. The fibers are
indeed very fine in that they approach the diameter of a single human hair. The
aromatic polyamide hollow fine fiber has an outside diameter of 85 micro-
meters for the brackish water fiber and an outside diameter of 95 micrometers
for the seawater fiber. Both of the fibers have an internal diameter of 42 micro-
meters.” The ratio of outside to inside diameter exceeds two. The fibers are
thick walled cylinders that have the compressive strength necessary to with-
stand the operating pressures. This concept is capable of achieving a high pack-
aging density of 3,000 square feet per cubic foot but the flux is considerably
lower than with sheet membrane.
Figure 4.11 is a schematic of a typical hollow fine fiber element. A con-
tinuous fiber is looped into a bundle around a central feed tube. The central
feed tube is sealed at the product water end and is perforated within the active
membrane area. Both ends of the fiber bundle are potted in an adhesive. When
cured, the adhesive and protruding fibers in the product water end are machined
EPW
Nub Shell
Hollow Fiber
Membrane Sheet Support
Bktck
Figure 4.11: Hollow fine fiber packaging.

Reverse Osmosis 281
to open the fibers at that end. The fiber bundle is then installed in a pressure
vessel and an O-ring in the product water end tube sheet seals against the pres-
sure vessel wail to prevent the high pressure reject stream from contaminating
the low pressure product stream. The element resembles a tube and shell heat
exchanger. The pressure vessels are fiberglass reinforced plastic and they are
designed and fabricated to accommodate operating pressures from 250 to
1,000 psig. Hollow fine fiber elements are available with nominal diameters
of 4,6 and 8 inches.
Pressurized feedwater enters the element through the central feed tube
which is connected to the high pressure feed manifold. The feed flows through
perforations in the tube and then radially between the spaces in the fiber toward
the outer diameter of the bundle. Longitudinal feed flow is minimized by the re-
ject end being potted. Product water permeates the membrane, enters the capil-
laries of the fiber bundle and flows to the tube sheet where it is discharged into
the product water plenum and then into the product water manifold. The reject
flows from the outer diameter of the fiber bund!es, through the end cap and to
waste via the high pressure reject manifolding. Normally, each hollow fine fiber
element is capable of recovering from 40 to 75% of the feed water from brack-
ish water and from 25 to 35% from seawater.
The advantages of the hollow fine fiber element are the high packaging
density and the elimination of membrane support materials. The prime disad-
vantage is the need for an efficient feedwater pretreatment to remove suspended
and colloidal solids.
Dynamic Membranes
Dynamic membranes are formed by flowing a feed solution containing 50
to 100 mg/!? of membrane-forming material tangential to a clean porous surface
at velocities from 5 to 50 feet per second under pressures from 500 to 1,200
psi.” The dynamic membrane is formed within minutes and the optimum
performance is usually achieved within an hour. The membrane-forming ma-
terials have been hydrous thorium, natural humic and fulvic acids, pulp mill
wastes, municipal wastes, synthetic polyelectrolytes, zirconium oxide, and poly-
acrylic acid. One of the most successful membrane-forming materials has been a
layer of polyacrylic acid deposited on top of a layer of zirconium oxide. The
porous surfaces used have been carbon, ceramic and stainless steel tubes. Use
of a stainless steel tube which had been pretreated with a filter aid and coated
with a layer of polyacrylic acid over a layer of zirconium oxide achieved a so-
dium chloride rejection of 90% with a flux of about 60 gfd. Recently, several
plants using dynamic membranes have been installed and there has been a re-
newed interest for such applications as orange juice concentration. These mem-
brane promise a high flux with a lower rejection and a low membrane cost.
However, the capital and operating costs have not always been commercially
competitive.
PLANT DESIGN
The first step in the design of an industrial reverse osmosis plant is to de-
termine the amount of water to be treated, peak demand, product water quality,
source of feedwater and reject discharge requirements. The next step is to ob-
tain an accurate and representative analysis of the feedwater. This analysis
should include determination of the concentration of the feedwater constituents
shown in Table 4.6. Knowledge of the feedwater constituents in Table 4.6 will
provide sufficient information for an experienced design engineer to successfully
design a reverse osmosis plant in the majority of applications.
Table 4.6: Feedwater Analysis
Cations Anions Other
Calcium Carbonate Silica
Magnesium Bicarbonate Total DissolvedSolids
Sodium Sulfate Total SuspendedSolids
Potassium Chloride Turbidity
Iron Nitrate Silt DensityIndex
Manganese Phosphate TemperatureRange
Ammonia Fluoride Total OrganicCarbon
PH
The above determinations with the exception of the silt density index should be
made in accordance with Standard Methods for the Examination of Water and
Wastewater. l3 A schematic diagram of the silt density index apparatus is shown
in Figure 4.12 and the silt density index (SDI) is determined as follows:
(I) Measure the amount of time required for 500 ml of feedwater to

flow through a 0.45 micrometer Millipore filter (47 mm in diam-
eter) at a pressure of 30 psig.
(2) Allow the feedwater to continue flowing at 30 psig applied pres-

sure and measure the time required for 500 ml to flow through the
filter after 5, 10 and 15 minutes.
(3) After completion of the test, calculate the SDI by using the equa-
tion below.
SDI = 100 (1 - Ti/Tf)

‘t
where SDI = Silt Density Index
Tt = Total elapsed test time (either 5, IO or 15 min-
utes)
Ti = Initial time in seconds required to collect the 500
ml sample
Tf = Time in seconds required to collect the second 500
ml sample after test time Tt (normally after 15
minutes).”
Reverse Osmosis 283
Qauge
at 3Opsig)
~illip~re Filter Holder
(0.46 mloron -47mm dla. Mlllipore FIlterI
Graduate Cylinder
(Measure Rate With Stop Watch)
Figure 4.12: Silt density index apparatus.
Manufacturers of hollow fine fiber elements usually require that the pre-
treated feedwater have an SDI of 3.0 or below in order for the element war-
ranty to be effective. One manufacturer of spiral wound element requires that
the pretreated feedwater have a turbidity of less than 1.O turbidity unit to main-
tain the element warranty, As a general rule, if hollow fine fiber elements are
to be used, the pretreated feedwater should have an SDI of 3.0 or less and, if
spiral wound elements are to be used, an SDI of 5.0 or less. However, spiral
wound elements have been used to recover municipal wastes with an SDI in
excess of 5.0 after pretreatment.
Pretreatment Section
The pretreatment section of a reverse osmosis plant is designed:
(1) to reduce suspended solids (1 micrometer or larger) to zero and to

minimize the effects of colloids in membrane fouling,
(2) to adjust the pH of the feedwater,
(3) to add a threshold inhibitor to the feedwater and
(4) to remove oxidizing compounds in the feedwater if required.
In some applications, such as the desalination of a well water, suspended

solids and colloids may be negligible and the pretreatment section may consist
of PH adjustment and addition of a threshold inhibitor. At the other extreme
is the desalination of surface waters, municipal wastes and industrial wastes
which may require all of the steps outlined above.
The following techniques can be used to remove suspended solids or to miti-
gate the effects of colloids:
(1) If the feedwater is a municipal water supply or well water with an
SDI of 6.0 or below it may only be necessary to filter the feed.
There are a number of filters available that have been successfully
employed and these are: diatomaceous earth, single media, dual
media and mixed media. A properly designed filter in these applica-
tions should be able to reduce the suspended solids load to zero.
(2) In-line coagulation may be used to reduce colloidal membrane
fouling. In this technique, coagulant aids such as alum, ferric chlo-
ride and/or a polyelectrolyte are added to the feed stream prior to
filtration. It is necessary to conduct on-site jar tests in order to
determine which coagulant aid is effective and what is the proper
dose for that coagulant aid.
(3) In the case of a feedwater with a high suspended solids, turbidity
and/or SDI, it may be necessary to use conventional coagulation
followed by sedimentation prior to filtration. Again, on-site jar
tests are required to determine the proper coagulant aid, dose and
settling rate.
The feedwater pH is usually adjusted to a pH of between 4 and 6 for the

following reasons:
(1) It is required (usually as a condition of warranty) to minimize the

rate of hydrolysis of the cellulose acetate ester. Cellulose acetate
hydrolysis reduces the useful life of the membrane by increasing
the flux and reducing the rejection of the membrane.
(2) The flux and rejection of some composite membranes are a func-
tion of pH and the optimum pH is between 4 and 6.
(3) Many natural waters are saturated in calcium carbonate which is
highly rejected by the membrane. Consequently, it is concentrated
in the feed/reject stream during the reverse osmosis process and it
will precipitate on the membrane decreasing the flux and rejec-
tion. Lowering the pH of the feedwater to between 4 and 6 con-
verts some of the carbonate or bicarbonate ions to carbon dioxide
and this prevents carbonate precipitation.
Some feedwaters contain compounds in addition to calcium carbonate that

may become saturated in the feed/reject stream and when that stream is concen-
trated in the reverse osmosis process, these compounds will precipitate on the
membrane with a resultant loss of membrane performance. Some of the more
sparingly soluble compounds of concern are:
Reverse Osmosis 285
Calcium sulfate Barium sulfate

Calcium phosphate Strontium sulfate
Calcium fluoride Silica
One method of preventing precipitation is to operate the reverse osmosis

unit at a recovery which will not concentrate the feed/reject stream to the com-
pound saturation level. Another method of preventing precipitation is to add a
threshold inhibitor, such as sodium hexametaphosphate, certain polyacrylates,
organophosphates or phosphonates, to the feedwater. The threshold inhibitors
are added at the rate of 1 to 5 mg/Q of feedwater, and they serve to disrupt the
formation of a crystalline precipitate during the residence time of the feedwater
in the reverse osmosis unit. In doing this, they broaden the solubility limits of
the sparingly soluble compounds.
As noted above, the polyamide and polyurea membranes cannot tolerate an
oxidizing agent, such as residual chlorine in the feed. Consequently, if these
membranes are used and the feed has a residual chlorine content, then it is
necessary to remove it. This is usually done by adding a stoichiometric excess of
sodium bisulfite or sodium thiosulfate in accordance with recommendations
from the membrane manufacturer.
Even with the best of pretreatment schemes, membrane elements will foul
over a long period of time and they must be cleaned. Each membrane packaging
configuration has a different degree of susceptibility to fouling and an ease of
cleaning.
The tubular element is the least susceptible to fouling and the easiest to
clean. This element is widely used in ultrafiltration applications where the proc-
ess streams contain suspended solids. The tubular element can be cleaned not
only by chemical action but also by mechanical means. A sponge rubber ball is
pumped through the tubular element with the chemical cleaning solution to
scour the membrane surface.
The spiral wound element requires less pretreatment than the hollow fine
fiber element or, stated in another manner, it is less susceptible to fouling. For
instance, the hollow fine fiber has been tested and found to foul excessively on
municipal wastewater reclamation applications while the spiral wound element
has been operated successfully. The spiral wound element must be cleaned with
chemical solutions, but there are no mechanical means available to clean this
element.
The hollow fine fiber element, with the great number of close packed fibers,
is an effective filter in itself. Consequently, it is the most easily fouled mem-
brane configuration and requires the most pretreatment. The hollow fine fiber
element can be cleaned with chemical cleaning solutions, but it is not amenable
to mechanical cleaning. It is also more difficult to clean than the spiral wound
element.
Element manufacturers have developed cleaning solutions for their specific
membranes and elements. Generally, there are two types of cleaning solutions-
one for removal of organic foulants and another for removal of metal hydrox-
ides. A cleaning system should be provided with each reverse osmosis system and
the cleaning system should consist of a tank to mix the cleaning solution and a
pump with associated piping, valves and instruments to circulate the cleaning
solution through the reverse osmosis elements.
Reverse Osmosis Section

Once the pretreatment study had been completed, it will be possible to de-
cide on the type of elements to be used in the reverse osmosis unit. If the SDI
of the pretreated feed is 3.0 or less, then either the spiral wound or hollow fine
fiber elements can be used. The choice will depend on economics (element
price) and desalination characteristics (flux and rejection). If the pretreated feed
SDI is more than 3.0, then the spiral wound element should be used. When the
decision as to element type is made, then it is appropriate to forward a copy of
the pretreated feed water analysis to reverse osmosis element manufacturers to
obtain a prediction of product water quality, recommended type of element,
total number of elements required, possible problems with sparingly soluble
compounds in the feedwater, allowable recovery, and price and delivery.
Figure 4.13 shows a flow diagram for a reverse osmosis unit with 75% re-
covery on a brackish feed. The pretreated feed is routed to the high pressure
pump where the feed pressure is raised to between 250 and 400 psig as required
for brackish water desalination. The pressurized feed is then pumped to the first
pass pressure vessels where about 50% of the feed .is recovered as product and
50% is reject. The reject from the first pass pressure vesselsis then routed to the
second pass pressure vessels where, again, about 50% of the first pass reject is
recovered as product and 50% is reject which is sent to waste. Thus, the over-
all recovery of the unit is 75% as product. As can be seen, a normal array for a
75% recovery unit is two first pass pressure vessels feeding one second pass
pressure vessel or a 2-I array. If the system recovery were from 40 to 60%,
all of the pressure vessels would be in parallel. However, if the system recovery
were raised to between 85 and 90%. the pressure vessels would be arranged in
a 4-2-l array.
pace”cT WATER - 250.000 QPO
+-A
Typl.Xl
SECOND PASS
REJECT
83.000 GPD
Figure 4.13: Reverse osmosis system.
The quality of the product water is a function of the rejection as shown in

the following equation:
Reverse Osmosis 287
cp = Cf * (1 - R)
where Cp = Concentration of the product

Cf = Concentration of the feed
R = Rejection
This equation produces accurate results for a membrane sample or a small ele-
ment with a low recovery, e.g., 2% or less. However, a practical reverse osmosis
system is designed to recover from 25 to 90% of the feedwater. This means that
the concentration of the feed varies throughout the membrane system. At 90%
recovery, the initial membranes will have a feed which is about IO times less
concentrated than the feed to the final membranes and the quality of the prod-
uct water will vary incrementally throughout the system. The product water
from the first membrane elements will be less concentrated than the product
water from the last elements. The product water from the practical reverse os-
mosis system is combined in the product water manifold and its concentration
is usually represented as the average product water concentration. The average
product water concentration is determined by the following formula:
i?$ = E”(l-R)
where cp = Average or total product water concentration

B = Average concentration of the feed and reject streams.
The average concentration of the feed and reject streams as represented by the
following equation is not truly representative:
7=~ = feed concentration + reject concentration

2
As Figure 4.13 shows, there are more elements in the first pass of a 75% re-
covery reverse osmosis system than in the second pass and the first pass elements
produce more product. Thus, a greater percentage of the total product water is
derived from the first pass and the total product water will be nearer in concen-
tration to the first pass average than to the second pass average.
Saltonstal114 has derived an equation to determine the average concentration
of the feed and reject streams accurately and this equation is as follows:
_ (1 - Y)(l
(1 - R) Y
- R)
1
where m, Cf and R are defined above and Y = system recovery.
The concentration of the total product then becomes:
cp = Cf *
1 - (1 - Y)(’
C (I - RI Y
- R, + (1 _ R)
1
Thus, the product water concentration is dependent on the feed concentra-

tion, membrane rejection and system recovery. Figure 4.14 shows the dependence
of product concentration on these variables. It is possible to use a lower rejec-
tion membrane and a high recovery with a low TDS feedwater. On the other
hand, a high TDS feedwater, such as seawater is limited to a low recovery and
requires a high rejection membrane.
500
_ 400
.
ol
E
;
t-
_ 300
;:
z 200
3
:
a
a 100
I I I I I
20 40 60 60 100
RECOVERY, %
Figure 4.14: Product water vs. recovery.
The membrane flux is also dependent on recovery. The average feed/reject

concentration increases with pressure and, since the osmotic pressure is propor-
tional to concentration, the average osmotic pressure will increase with increased
recovery. At constant applied pressure, the membrane flux will decrease with in-
creased recovery. Figure 4.15 shows the increase in average osmotic pressure and
the decrease in flux for a membrane which would produce 20 gfd at zero recov-
ery operating on a feed of 5,000 mg/Q at 450 psi applied pressure. Concentration
polarization is assumed to be constant at all recoveries.
Reverse Osmosis
26 ‘160
FEED - 6.000; m2ll
MEMBRANE REJECTION - 96%

NET PRESSURE - 400 PSI
20 I- -120
16 *QO
10 -80
6 -30
0 *o
I I I I
O 20 40 60 SO 1 0
RECOVERY, %
Figure 4.15: Flux and average osmotic pressure vs. recovery.
It is noted that the osmotic pressure of seawater is about 400 psig and an
operating pressure of between 800 and 1,000 psig is required to obtain optimum
flux and rejection. If a recovery of only 50% were used, the reject osmotic pres-
sure would be in excess of 700 psig and the average osmotic pressure would be
about 550 psig. This would present problems in obtaining a potable product of
500 mg/Q or less. Consequently, most seawater desalination reverse osmosis units
are operated at a recovery of from 25 to 35%.
Since lowering of the feed pH usually means that the feed contains signifi-
cant quantities of carbon dioxide, the feed and reject streams are extremely cor-
rosive. In addition, the membrane does not reject carbon dioxide and, conse-
quently, the product stream contains about the same concentration of carbon
dioxide as the feed and reject streams, i.e., all three process streams are corro-
sive. General practice is to use PVC pipe and valves for low pressure piping (less
than 100 psig) and 316 stainless steel pipe and valves for the high pressure
(above 100 psig) process streams. The high pressure pump wetted parts should
be either 316 stainless steel, corrosion resistant plastic or other corrosion resis-
tant materials. The pressure vessels are fabricated with fiberglass reinforced plas-
tic which is corrosion resistant.
The reverse osmosis process is relatively simple and instrumentation re-
quirements are minimal. Following is a list of the minimum recommended para-
meters to be measured in a reverse osmosis system:
Feed Product Other
FlOW Flow Differential Pressure
Temperature Conductivity across each array
Pressure
Conductivity
PH
A variety of control schemes can be incorporated in the design of a reverse

osmosis plant. However, this subject is beyond the scope of this manual.
Posttreatment Section
In most plants that use reverse osmosis in the preparation of process water,
the reject stream is routed directly to waste discharge without any additional
posttreatment. In industrial plants that use reverse osmosis to treat industrial
wastes, the reject stream may contain valuable materials and this stream would
be sent back to the process. In other applications of industrial waste treatment
by reverse osmosis, the reject stream may require additional treatment prior to
ultimate discharge. In this case, the reverse osmosis unit will have provided a
large volume of water that is disposable or can be reused (the product) and a
smaller volume of reject which can be treated more economically.
The product water from a reverse osmosis unit will have a low pH and most
probably a high concentration of carbon dioxide. The carbon dioxide can be
removed and the pH of the product increased by use of a decarbonator. A de-
carbonator is a packed column in which product water is introduced at the top
while either forced or induced air is introduced at the bottom. The air and water
flow countercurrently over and around the column packing. The carbon dioxide
is stripped from the water and exits from the decarbonator at the top in the air
stream. In a well-designed decarbonator, the carbon dioxide content can be re-
duced to about 5 mg/Q in the water effluent.
INDUSTRIAL REVERSE OSMOSIS AT A REFINERY
One of the most innovative industrial uses of reverse osmosis is at the

Petromin Refinery in Riyadh, Saudi Arabia. The refinery takes an unusable
municipal wastewater, secondary effluent from the Riyadh sewage treatment
plant, and by using lime clarification, filtration, reverse osmosis and ion ex-
change demineralization, it converts that useless waste into the entire process
water requirements for the refinery. Figure 4.16 is the process flow schematic
for the refinery water treatment plant.
Chlorinated secondary effluent arrives at the pumping station, which is
adjacent to the sewage treatment plant in Riyadh, through an open channel
which flows to the pump station inlet basins. The 4.63 MGD of effluent is then
pumped 19 kilometers to the refinery by either of two full capacity pumps
which take suction from the inlet basins. The effluent is pumped through a
2
2 Sludge Thickener Overflow
t
Slaked Lime
: Ion Exchange Regenerant
2 t 9
Iii
r
d
I I
I 0.35 MOD
c
RAPID MIX, COAGULATION
t
:
Eo,_
4
,
SURGE PONDS ’
8.04
1.500
MGD
mgli
+ 6 CLARIFICATION 5.02
1,100
MGD
mgli
.
$z “E Plant Use Return
Sludge to Thickener
0 m)o” 4
1.08 MOD D
= ?a Backwash 0.37 MOD
: 0 d I
0.13 MOD
rz
c I
PRIMARY
Reject
REVERSE OSMOSIS
1.12 MGD- FILTRATION
SYSTEM 4.27 MOD * 5.48 MOD
h
1
z A
2
i ‘g’p;ke
:
P
4 : General Plant Use’
z
L SO mg/iv 0 : d 1.08 MOD
; gz Regenerant
SECONDARY
0 I; 0.02 MGD + :
+ REVERSE OSMOSIS. IL
1.13 MOD )
z0 (0 SYSTEM z ;
z, . 10 mgli 0 =
= B ION EXCHANGE
mo
: D
or 1.12 MOD SYSTEM
0
0 - 10 mg/i
I-
Figure 4.16: Industrial reverse osmosis at a refinery.

24:inch diameter cement lined carbon steel pipeline. The pipeline is equipped
with vacuum relief valves and air-release safety valves at the two highest Points
ln the line plus air-release safety valves at the third highest point in the line.
The pipeline is also equipped with pigging facilities (for pipeline cleaning) which
consist of 2 pigs, a pig launcher and pig receiver.
On arrival at the refinery, the effluent is discharged into a 2 compartment
splitting box which divides the flow evenly for discharge into the two on-site
surge ponds. The effluent flows from the splitting box by gravity into the surge
ponds. At the same time, miscellaneous refinery plant streams and backwash
from the refinery water treatment plant filter are returned to the surge ponds
and mixed with the secondary effluent. The return stream flow is estimated to
be 1.19 MGD with a quality similar to the secondary effluent. Each of the surge
ponds is a concrete lined, earth basin with a capacity of 5.3 million gallons. The
surge ponds have a static air diffusion system to prevent sewage septicity and im-
prove homogeneity in the secondary effluent. Chlorine can be added to the sec-
ondary effluent either at the sewage treatment plant pump station or at the
surge ponds at the refinery, if required.
All surface waters and municipal effluents contain suspended solids as well
as dissolved solids and the presence of suspended solids dictates the need for a
pretreatment section. Experience has shown that effective removal of the sus-
pended solids in pretreatment is a prerequisite to efficient reverse osmosis mem-
brane performance. Suspended solids in secondary effluent are primarily organic
in nature and, due to their small size, it is difficult to remove them by settling.
Therefore, it is necessary to aggregate the smaller particles into larger particles
which can more easily be removed by settling and filtration.
For some time, it has been known that suspended solids or colloids in water
possessan electrical charge which is predominantly negative. As a result of this
electrical charge, the colloidal particles tend to repel each other and thereby
limit the potential for aggregation and the colloidal suspension is said to be sta-
bilized. Before the SLQ ended particles can coalesce, the stabilizing forces must
be neutralized and this is usually done by adding chemicals. Trivalent aluminum,
ferric hydroxides, and/or lime have been used to destabilize colloidal suspensions
and allow suspended particles to grow in size. Another means of particle destabi-
lization is the use of natural or synthetic polymers which are long chain, high
molecular weight polyelectrolytes with many active sites. The negatively charged
colloids are adsorbed on the active sites with the resultant coagulation or growth
of the particles.
The refinery clarification equipment has the capability of adding any of the
chemicals mentioned above. However, lime clarification was chosen as the method
to be used. The combined secondary effluent and the plant return streams (5.64
MGD) are pumped from the surge ponds to the rapid mix basin in the clarifier.
The rapid mix basin has two compartments in series and each compartment has
a high speed mixer. Lime and sodium hydroxide are added to the first compart-
ment and these are vigorously mixed with the secondary effluent in both com-
partments. As a result, the pH of the effluent from the rapid mix basin is raised
to between 10.8 to 11 .O. At this pH, much of the bicarbonate in the water re-
acts with the lime and forms an insoluble calcium carbonate and the magnesium
in the water reacts with hydroxyl ions to form insoluble magnesium hydroxide.
The water flows from the rapid mix basin to the flocculation basin which
has four compartments. Each compartment is further divided into three sec-
Reverse Osmosis 293
tions and all twelve sections are equipped with a slow speed turbine mechanism.
The gentle agitation provided in the flocculation basin promotes contact be-
tween the calcium carbonate, magnesium hydroxide and suspended matter
which results in the formation of larger particles.
The water then flows by gravity from the flocculation basin to the sedimen-
or clarification basin. The enlarged floe particles settle to the bottom of this
basin and they are removed as a 1% sludge by three circular rakes into a sludge
basin or hopper. This amounts to about 0.37 MGD of sludge which is pumped to
the sludge thickener tank. The sludge is thickened to about 6% in the sludge
thickener and then filtered in a plate and frame filter press to about 50% solids.
The filtered sludge may then be transported to a a landfill for disposal or sent to
the on-site multiple hearth lime kiln for regeneration. Operation of the lime kiln
is economically marginal. The overflow from the sludge thickener and the filter
press filtrate are returned to the rapid mix basin for reclamation. In addition,
part of the water used in the regeneration of the ion exchange demineralizers is
also returned to the rapid mix basin for recovery. The total amount of water
from these three sources is 0.35 MGD with variable quality.
Effluent from the clarifier is saturated in calcium carbonate and this would
precipitate on the filter media to clog the filter which is the next step of pre-
treatment. Consequently, the clarifier effluent flows by gravity to the serpentine
recarbonator basin where carbon dioxide is added to reduce the pH to between
7.5 and 8.0. The insoluble calcium carbonate and magnesium hydroxide are con-
verted to soluble calcium and magnesium bicarbonate in the recarbonator basin.
The 5.62 MGD of effluent from this basin has a suspended solids concentration
of about 2 mg/R and a TDS of about 1,100 mg/Q.
Following lime clarification and recarbonation, the treated water is pumped
to the fire water storage tanks (0.14 MGD) and the remainder (5.48 MGD) is
pumped to the water treatment plant cooling tower. The cooling tower is used
to reduce reverse osmosis feedwater temperature when required. For a large
part of the year, the cooling tower is inoperative. A static mixer is installed in
the line upstream of the cooling tower. At this point, sulfuric acid can be added
to the process stream to reduce the pH of the water in the event that carbon di-
oxide is unavailable to the recarbonation system. Chlorine can be added to the
process water at the cooling tower influent line or at the chlorine diffuser of the
filter aid basin. The water flows by gravity from the cooling tower to the filter
aid mix station where it is possible to add coagulant aids to assist in the removal
of suspended solids during filtration. Coagulant aids are not being added at the
filter aid mix basin at this time. The water then flows by gravity to the dual
media filters.
There are four separate filter basins with filter media as described in Table
4.7 below.
Table 4.7: Filter Media Description

Height
Media mm Inches Size, mm

Coal so0 31.5 0.6 - 1.6
Sand 400 15.7 0.6 - 1.0
Gravel Support 400 3 - 35

The process water flows through the filter media by gravity to remove the re-
maining suspended solids. It then flows through a flow control valve to the fil-
tered water reservoir by gravity. Backwash water overflows from the filtered
water reservoir to the wash water storage tank. The filter effluent of 5.48 MGD
provides 1.06 MGD of general use water to the refinery, 0.13 MGD of filter
backwash water, 0.02 MGD of filtered water to the lime slakers at the rapid mix
basin and the remaining 4.27 MGD flows by gravity to the reverse osmosis de-
chlorination basin.
Chlorine has been added to the feedwater upstream of reverse osmosis pre-
treatment. However, since chlorine will depolymerize the polyurea membrane
barrier layer in the spiral wound element, with subsequent loss of desalination
properties, the chlorine is removed in the pretreatment system dechlorination
basin. This removal is chemically accomplished by the addition of sodium bi-
sulfite. The chlorine level in the influent and effluent to the dechlorination basin
is continuously monitored. The feedwater is then transferred from the dechlor-
ination basin to the cartridge filter feed pumping station by gravity flow and it
is then pumped to the cartridge filters.
Sulfuric acid and sodium hexametaphosphate (SHMP) are injected in the
feedwater line upstream of the cartridge filters. The sulfuric acid is injected to
adjust the feedwater pH to a level of between 4 and 6. The purpose of acid in-
jection is twofold. The primary purpose is to mitigate the possibility of calcium
carbonate deposition by conversion of bicarbonate to carbon dioxide. Coinci-
dentally, the rejection performance of the thin film composite membrane is pH
sensitive and the optimum performance is at the operating pH level. SHMP is
added to the feedwater as a threshold inhibitor to inhibit the crystalline growth
of sparingly soluble salts such as calcium sulfate.
After chemical addition, the feedwater is routed to cartridge filters which
serve to mix the chemicals which have been added upstream and to insure that
any particles that may have escaped the gravity filters, such as sand or other par-
ticulate matter is removed. In general, the cartridge filters do not improve the
quality of the reverse osmosis feedwater to a large degree and they are not in-
tended as continuous duty filters. The effluent from the cartridge filters is
routed to the primary reverse osmosis feed pump wet well.
The primary reverse osmosis system contains five trains of equal capacity.
Each train is operated independently and there is a vertical turbine high pressure
pump for each train mounted on the primary feed pump wet well. The high pres-
sure feedwater is pumped to the reverse osmosis train where the operating pres-
sure of each train is adjusted by the pressure control valve upstream of the spiral
wound element pressure vessels.The elements are located in fiberglass reinforced
plastic pressure vessels with six elements per pressure vessel. The elements have
nominal dimensions of 40 inches in length by eight inches in diameter and they
contain composite membrane with a polyurea membrane barrier layer. Each pri-
mary train has 180elementsin 30 pressurevesselswhichare manifolded in a20-10
array. The pressurized water is fed to 20 pressure vesselsin parallel where about
50% of the feed is recovered as product and the reject from these pressure ves-
sels is manifolded and fed to 10 pressure vessels in parallel. About 50% of the
feed to these latter pressure vesselsis recovered as product and the reject is man-
ifolded through the flow control valve to waste. The product water from the
primary reverse osmosis system is transferred to the cooling tower makeup and
Reverse Osmosis 295
desalt water storage tanks or to the secondary reverse osmosis system feed pump
wet well. The overall recovery for each train is 75% and the design capacity of
each train is 0.83 MGD of product for a total product water capacity of 4.10
MGD in the first stage. The daily requirements for product water are 3.35 MGD
so there is an installed spare capacity of about 20%. The product water require-
ments are 2.02 MGD for the refinery cooling towers and the desalter and 1.33
MGD for the second stage reverse osmosis system. The reject (1.12 MGD) from
the first stage reverse osmosis system is sent to the on-site solar evaporation
pond for disposal.
The first stage product water concentration of 80 mg/Q is adequate for cool-
ing tower makeup and desalter process water but it is not pure enough for mod-
erate pressure boiler feed. Consequently, the first stage product must be further
treated with ion exchange demineralization to achieve the desired purity. A sec-
ond stage of reverse osmosis is used at the refinery to pretreat the first stage
product prior to ion exchange demineralization. An improved quality feed to the
ion exchangers will improve not only the quality of ion exchanger effluent but it
will reduce the quantity of regenerative chemicals required by the ion exchangers.
The secondary reverse osmosis system contains three trains of equal capac-
ity. As in the primary reverse osmosis system, each train is operatred independ-
ently and there is a vertical turbine high pressure pump associated with each
train. The pressurized feedwater is pumped to the secondary trains where the
operating pressure of each train is adjusted by the pressure control valve up
stream of the element pressure vessels which are fabricated from fiberglass rein-
forced plastic. Each secondary train has 21 pressure vessels (126 elements)
which are manifolded in a 12-6-3 array. The spiral wound elements in the sec-
ondary trains are identical to the elements in the primary stage. The pressurized
feedwater is fed to the first 12 pressure vessels in parallel where about 50% of
the feed is recovered as product and the remaining 50% is manifolded and fed
to the next six pressure vessels in parallel. Product recovery in these pressure
vessels is again about 50% and the reject is manifolded and fed to the last three
pressure vessels where about 40% of the feed is recovered as product. The re-
ject from the last three pressure vessels is manifolded through the flow control
valve in each train to the dechlorination basin. The reject flow of 0.20 MGD
from the secondary reverse osmosis trains is of better quality than the feed to
the primary reverse osmosis system and, thus, it is used as feed to the primary
system. The overall design recovery of each of the secondary trains is 85%. Each
of the trains is rated at 0.57 MGD for a total capacity of 1.71 MGD. The total
required capacity is 1.31 MGD which can be produced by two trains with one
train in standby. The recovered product is routed to the forced air decarbonators
in the ion exchange demineralization system.
The product water from the secondary reverse osmosis system contains a
high concentration of carbon dioxide as a result of pH adjustment in the pri-
mary reverse osmosis system which converts bicarbonate and carbonate alkalin-
ity to carbon dioxide. Since the membrane is “transparent” to carbon dioxide,
it passes through both the primary and secondary reverse osmosis systems into
the secondary system product water. Although the ion exchange resin would
remove the carbon dioxide, it is more economical to do so in the decarbonator.
There are two degasifiers (decarbonators) in the refinery water treatment
plant and each of these are in parallel and normally in operation, i.e., one de-
gasifier is not used as a standby. The secondary product is fed to the top of the
degasifiers where it is allowed to cascade over the degasifier packing. Air is
blown through the packed section from the bottom and it rises in the packed
column countercurrent to the water. It is estimated that 0.01 MGD of product
water are lost to evaporation in the decarbonator. Since the water is at a low pH,
carbon dioxide is transferred from the liquid phase to the gas phase. The result
is that the carbon dioxide concentration in the water is reduced to about 5 mg/ll.
Two transfer pumps, both full capacity, are installed to pump the decarbonated
water through the ion exchange system.
There are two full capacity ion exchange trains in the system and each train
consists of a cation tank, an anion tank, instrumentation, and the necessary
valves and piping for process control. The ion exchange demineralization is ac-
complished in a two-step process involving treatment with both cation and anion
resins in separate process vessels. The water is first passed through the strong
acid cation exchange resin (Amberlite IR-20) to exchange the cations for the hy-
drogen ion. The effluent is then passed through a Type I, strong base anion ex-
change resin (Amberlite IRA-410) where the anions are exchanged for the hy-
droxide ion. The result is that cations and anions are substituted by water mol-
ecules and a high purity effluent is available for use as boiler feedwater. When
the ion exchange capacity of one train is depleted, this train is removed from
service, the standby train is placed in operation and the depleted train is regen-
erated with acid and caustic. About 0.02 MGD of decarbonator effluent are used
for regeneration of the resins and much of this is returned to the rapid mix ba-
sins for recovery.
REVERSE OSMOSIS AND ION EXCHANGE
The preceding example of a reverse osmosis industrial application at a re-

finery showed that the process is capable of:
(1) treating a feedwater with high suspended solids and dissolved sol-
ids concentrations;
(2) reclaiming a water that is considered by most as unusable; and
(3) developing a number of process streams with different quality
requirements.
Reverse osmosis also has been used to treat municipal water supplies for in-
dustrial purposes even though these supplies are generally low in turbidity, sus-
pended solids and dissolved solids. A large number of reverse osmosis systems
have been installed in industrial plants to prepare industrial process water with
municipal water as the feed source. A significant number of these industrial ap-
plications are to either replace ion exchange demineralization or to pretreat
municipal supplies prior to ion exchange demineralization.
Reverse osmosis systems now commercially available will remove 95% or
more of the dissolved solids normally removed in ion exchange, and as will be
discussed later, a few that are not. The ionized solutes are not all removed to
the same degree by reverse osmosis any more than ion exchange resins have the
same effect on all solutes. Divalent and multivalent ions, such as calcium, mag
nesium, sulfate, iron and manganese, can be rejected to greater than 99%. So-
Reverse Osmosis 297
dium, potassium and chloride are normally rejected to the 95% level or better.
The net effect is to reduce the number of regenerations required of the ion ex-
change columns by a factor of 20 or more. This results in a significant reduc-
tion in the amount of waste regenerant solutions that must be disposed of and
a material reduction in the dissolved solids that might normally be discharged
to the environment. A concentrate or reject is produced by reverse osmosis,
but there is little change in the environmental salt budget. Concommitant re-
sults are a major decrease in the space and equipment necessary for regenerant
storage and an extension of the useful life of the resins, owing to reduced resin
attrition. In addition, substances difficult to remove from the resin, which also
affect its performance, are greatly reduced or are removed by reverse osmosis
pretreatment.
When reverse osmosis is used for preliminary demineralization, the variation
in the quality of the ion exchange demineralized water is reduced. The amount
of dissolved solids in the feed to the ion exchange beds is 5% or less than when
the raw water is fed directly. It is, therefore, obvious that the variation in solids
in the finished water will also be less when breakthrough occurs. As a result,
the resins may be utilized more efficiently. Additional reliability and control
can be gained by measuring the solids concentration following reverse osmosis
treatment. This precaution virtually eliminates shock loadings on mixed bed
polishing columns. Where small point of use polishing columns are used, such
as in microelectronics manufacturing, the danger of a rapid breakthrough be-
comes considerably reduced, and there is an improvement in production and
product quality.
An important factor to be remembered is that in some cases water supplies
unsatisfactory for processing to high purity water may be the only sources avail-
able. Preliminary demineralization by reverse osmosis will make this water suit-
able for subsequent demineralization by ion exchange. It is thus apparent that
such an economically important factor as plant site location, which may be de-
pendent on the availability of suitable water, can be made more flexible through
the use of reverse osmosis. It may be possible now to utilize seawater as a source
of industrial process water.
During the 1960’s, reverse osmosis was compared with other methods of
demineralization. It was indicated in these comparisons that reverse osmosis
could not compete favorably with ion exchange at dissolved solids concentra-
tions below 700 mg/P and that its most favorable area of use would be from
about 1,200 to 5,000 mg/Q dissolved solids. This idea has been totally refuted
because some of the most successful applications of reverse osmosis, particularly
as part of the process to produce high purity water, have been in treating low
dissolved solids water. Water containing 200 mg/Q dissolved solids or less has
been treated at costs equal to or lower than those of ion exchange alone.
Two substances that are frequently of concern in ion exchange demineraliz-
ation are silica and organics. The organics are frequently present in natural wa-
ters as aromatic polycarboxylic acid derivatives known as humic and fulvic acids.
Silica may be the limiting factor in the efficiency of the anionic resins, and (par-
ticularly in boiler feedwater applications) the lower the concentration before
ion exchange demineralization, the better. Reverse osmosis will frequently pro-
duce 90% or greater reductions in total silica concentrations. However, perform-
ance should be tested on the specific water to be treated since me results can be
variable and the reason for differences between waters is not yet understood.
Where the silica concentrations in the raw water are high, reverse osmosis has
been most effective. Even in trace quantities, humic and fulvic acids have been
responsible for impairing the life of anionic resins and affecting performance of
ion exchange columns. These organics are readily removed by reverse osmosis
membranes.
REVERSE OSMOSIS AND POLLUTION CONTROL
Pollution control legislation has made industry aware of the economic pen-
alty for inefficient use of raw materials. The plating industry was one of the
first industries to experience this enforced awareness. Loss of raw materials in
this industry” can result in five distinct costs:
(1) replacement of the plated material that was discharged to waste

prior to initiation of pollution control practices;
(2) removal of that material from the wastewater after pollution con-
trol enforcement;
(3) disposal of the residue from item 2;
(4) replacing of the process water (at times quite expensive) lost in
wastewater; and
(5) treatment of the wastewater in a publicly owned treatment works
after discharge into a sewage system.
In response to the plating industry’s increased awareness of the above costs,

plating shops have modified their practices to reduce their material and eco-
nomic losses. While pollution control was greeted by many with coolness, the
resultant cost savings have shown that pollution control is not as onerous as ex-
petted .
Reverse osmosis has been installed in many plating shops as a way to re-
source recovery and minimizing the size of waste treatment equipment and vol-
ume. Reverse osmosis is particularly suited to waste treatment in the plating in-
dustry because most of the toxic ions in the plating solutions are well rejected
by commercially available membranes. Table 4.8 shows the rejection ranges of
some of the more common toxic materials in the plating industry.16
Table 4.8: Reverse Osmosis Rejection Range
-Ion RejectionRange, X
Nickel 98 - 99
Cower 98 - 99
Cadmium 96 - 98
Chromate 90 - 98
Cyanide 90 - 95
Zinc 98 - 99
Reverse Osmosis 299
The nickel plating industry is a typical candidate for the use of reverse os-
mosis in pollution control. Figure 4.17 shows a schematic of this industrial ap-
plication. The workpiece travels from the plating bath with a concentration of
270,000 mg/Q to the rinse tanks. There are three rinse tanks in series and rinse
water flows countercurrent to the workpiece. The work piece drags out plating
bath to the first rinse tank, first rinse tank solution to the second rinse tank and
second rinse tank solution to the third rinse tank. Consequently, the first, sec-
ond and third rinse tanks have concentrations of 3,000 mg/Q, 333 mg/ll and 37
mg/ll, respectively.
About 100 gallons/hour (GPH) are pumped from the first rinse tank through
a cartridge filter and into a reverse osmosis unit. The reject stream contains 99%
(59,400 mg/Q) of the nickel in the feed stream with 1% (32 mg/Q) remaining in
the product stream, The reject stream is routed through an activated carbon col-
umn to the plating bath. The reverse osmosis product stream is combined with
5 GPH of tap water makeup, which is added to compensate for surface evapora-
tion in the plating tank, and the combined stream is returned to second rinse
tank. The waste stream (10 GPH) is sent to waste treatment which is a precipi-
tation process.
WORKPIECE
Figure 4.17: Reverse osmosis in nickel plating.
REVERSE OSMOSIS AND SEAWATER DESALINATION
It has been estimated that the oceans cover about 70% of the earth’s area
and contain about 80% of the water on or in the earth. Seawater concentra-
tion varies from 25,000 mg/!.? of total dissolved solids up to over 50,000 mg/ll
in the Arabian Gulf. The average composition of seawater is about 35,000 mg/lZ
with the major ions being shown in Table 4.9.
Table 4.9: Major Constituents of Seawater
Constituent Concentration, mg/l
Sodium 10,500
Magnesium 1,350
Calcium 400
Potassium 380
Chloride 19,000
Sulfate 2,700
Bicarbonate 142
Bromide 65
Other Ions 33
Total 34,570
While seawater is abundant, it is to a large degree unusable and it is the con-

stant dream of the desalination industry to desalt seawater in an economical
manner that would allow the product to be used for agricultural purposes. This
goal is a long way from reality. On the other hand, distillation and now reverse
osmosis are desalting seawater efficiently enough to be used for preparing pot-
able waters in affluent areas. Seawater, with a concentration of 35,000 mg/P, has
an osmotic pressure of almost 400 psig and this mandates a membrane element
with high pressure capabilities. If the reverse osmosis seawater plant were re-
quired to deliver a product water of potable quality (500 mg/Q or less), then the
membrane element would require a minimum rejection of 98.6% to attain that
product quality at 0% recovery. At higher recoveries, both pressure and rejec-
tion must increase to obtain potable quality. Spiral wound and hollow fine fiber
elements have been developed with the capability of operating at 800 to 1,000
psig and sodium chloride rejections up to 99.5%. This capability has placed re-
verse osmosis in the seawater desalination business and reverse osmosis is be-
ginning to replace distillation as the result of the clear cut economic advantage
of reverse osmosis.
An example of reverse osmosis seawater desalination for industrial purposes
is the system installed in a thermoelectric power plant in Venezuela in 1980.”
The original segment of the plant is designed to produce 800,000 GPO of boiler
feedwater and potable water. A process flow diagram for this system is shown in
Figure 4.18.
Seawater is pumped from an intake channel to the roughing filters which
are the first part of the pretreatment process. Ferrous chloride and chlorine
are added to the filter influent line. The chlorine disinfects the seawater and oxi-
dizes the ferrous ion to the ferric ion which forms insoluble ferric hydroxide.
The ferric hydroxide acts as a coagulant aid. A large Percentage of the coagu-
lated particles are removed in the roughing filter which reduces the feedwater
Sol from 15 to about 3. There are five gravity roughing filters which have a dual
media of sand and anthracite. Four of the filters are in operation while one iS
Reverse Osmosis 301
in standby to be placed in operation when one of the operating filters is back-

washed.
The effluent from the roughing filter clearwell is pumped to the pressurized
polishing filters which further reduce the SDI to less than 3. There are four op-
erating polishing filters and one standby filter. The filter media is anthracite
and green sand.
Sodium bisulfite is added to the effluent from the polishing filters to re-
move residual chlorine which would be harmful to the polyamide reverse os-
mosis membrane. Sulfuric acid is then added to the filter effluent to adjust the
pH to about 6.5 and the pretreated water is routed to the primary reverse os-
mosis system.
The first stage reverse osmosis systems consist of four trains which are cap-
able of producing 200,000 GPD each of product water. Each train consists of a
5-g cartridge filter with a high pressure pump feeding two subunits of 25 DuPont
B-10 permeators each plus the necessary piping, valves and instrumentation. The
first stage system reverse osmosis units are operated at 30% recovery with a
pump discharge pressure of 900 psig. The total dissolved solids content of the
feed was reported to be 37,000 mglR and the product concentration from this
stage was 490 mg/Q. This product water is sent to a 262,000-gallon storage
tank which provides feed to the two second stage reverse osmosis trains and it
is the source of potable water for the power plant. The reject from the first
stage is routed to waste. A high pressure recovery turbine is not used and a con-
siderable amount of energy is wasted.
From Seawater Intake
I FILTERS
I FIRST
REVERSE
STAGE
OSMOSIS
I
ti
3.407 kgpd
30% RECOVERY
37.000mg11
; 800 kgpd
: 4gOmgll
0
-
a.
SECOND STAGE
Potable Water
I
REVERSE OSMOStS c;
133 kggd
867 kgpd
85% RECOVERY
HOLDING 490 mgll
4gOmgtt
TANK
Figure 4.18: Power plant process water from seawater.

The first stage product quality meets potable standards but it is far from
the quality required for boiler feedwater. Preparation of boiler feedwater is done
in an ion exchange system and, with an influent of 490 mg/Q, the operating cycle
would be short and chemical regenerant cost high. In this case, it is economical
to further treat the first stage reverse osmosis product in a second stage reverse
osmosis system to further decrease the load on the ion exchange resins.
About 85% of the first stage reverse osmosis system product water is pumped
to the second stage reverse osmosis system which consists of two trains. Each
train consists of a 5-p cartridge filter, a high pressure pump, DuPont 8-9 per-
meators, piping, valves and instrumentation. Each train has 22 permeators in a
12-7-3 array and it is operated at 85% recovery. The product from the second
stage has a TDS of 28 mg/ll and this is sent to the ion exchange system. The re-
ject from the second stage is of much better quality than the incoming sea-
water and this is routed to the roughing filter clearwell.
The reverse osmosis system has operated well and the capacity of the plant
has been increased to twice that of the initial segment described above.
GENERAL APPLICATIONS OF REVERSE OSMOSIS
The above applications were specific examples that were chosen to demon-
strate the versatility of the reverse osmosis process. The following list enumer-
ates a wide variety of applications for which the reverse osmosis process may be
considered:
(1) Municipal Potable Water

General quality improvement of present supplies
Upgrade total municipal supply
Potable water from degraded supplies
Brackish water desalination
Seawater desalination
Removal of nitrates, fluorides, heavy metals, etc.
Bottled water production
(2) Industrial Water
Provide usable water where none available
Brackish water desalination
Seawater desalination
Pure water production
Industrial rinse waters
Food industry
Electroplating
Power plant boiler feed
Beverage production
Medical
Ultrapure water production
Pharmaceutical
Reverse Osmosis 303
Electronics
Medical
(3) Municipal Wastes
Reclaim municipal wastewaters (sewage) for
Ground water recharging
Agricultural or landscape irrigation
Industrial process water
Improve effluent quality to meet discharge requirements
(4) industrial Wastes
Reclaim industrial wastewaters for
Reuse within industrial plant
Zero discharge
Agricultural or landscape irrigation
Removal of toxic substances prior to discharge
Resource recovery
Separate or concentrate valuable materials
Electroplating industry
Dairy industry
(5) Miscellaneous
Production of pure water for high value crops
Recovery of agricultural irrigation drainage
Production of shipboard drinking water
COSTS OF REVERSE OSMOSIS
There is no such thing as the typical cost of a reverse osmosis system or

of the product water from that system as these costs depend on a number of
things:
Economic conditions Power costs

Market conditions Site preparation
Plant size Product quality requirements
Local labor rates Feedwater availability
Chemical costs Accounting procedures
The DuPont Permasep Engineering Manual” has published a “guide” for the
capital, operating and maintenance costs for both a brackish water system and a
seawater system. The brackish water system costs are shown in Table 4.10. They
are based on a large brackish water system built in the southern United States in
1982. The estimated capital cost of the plant is $1.25 per gallon per day of prod-
uct water installed. This cost includes the cost of wells, a reverse osmosis system
with pretreatment, a building for the reverse osmosis systems and office. The
above installed capacity cost does not include the cost of land nor an independ-
ent power source.
Table 4.10: Total Water Cost for Brackish Water RO
Cost/l,000 Gals. Product
Energy (O.O6/KWH) $0.36
Chemicals 0.09
Labor 0.12
Maintenance and Repair 0.05
Membrane Replacement 0.10
Amortization (12%/20 years) 0.48
Total $1.20
Data are provided for a seawater system that will produce IO million gallons
per day of product water. The system is a two stage system similar to the one
shown in Figure 4.18 where the product water from the first stage high pressure
seawater system is treated in a second stage lower pressure brackish water sys-
tem. The product water TDS is 200 mg/g. The estimated 1982 cost for such a
system was $45 million which includes the RO system with pretreatment and a
building for the RO system, controls and office. It does not include the cost of
land or an independent energy source. The total water cost for such a plant is
shown in Table 4.11.
Table 4.11: Seawater RO System
Cost/l,000 Gals.Product
Energy ($O.O6/KWH) $1.80
Chemicals 0.14
Labor 0.19
Maintenance 0.22
Membrane Replacement 0.90
Amortization (12X/20 years) d1 75

Total $5.00
Both the brackish and seawater reverse osmosis product water costs are
based on 1982 costs and they are indicative of specific plants in an assumed
location in the southern United States. The cost of energy in the seawater sys-
tem assumes that the reject from the first stage high pressure reverse osmosis
system is sent to an energy recovery system which reduces the overall energy
requirements for the total system by 31%.
Reverse Osmosis 305
FUTURE PROJECTIONS
The initial projections of 20 years ago have proven to be unrealistic in that

reverse osmosis has not caused deserts to bloom, nor does every household con-
tain a reverse osmosis unit to improve the tap water. Yet, the process has been of
economic value in providing process water to industry, potable water to high in-
come arid regions and a method of reclaiming municipal and industrial wastes.
As of 1985, it was estimated that the worldwide market for reverse osmosis
membrane elements (not total systems) was about $50 million.
The future of the process rests in the research that will result in the develop-
ment of a product that will compete with and win over other separation proc-
esses that can do the same thing. Reverse osmosis has gone a long way toward
becoming more economical than multistage flash distillation in the production
of potable water from seawater. The development of membranes with higher
fluxes and improved rejections will be incorporated into single stage seawater
plants with lower operating pressures. This will not only permit reverse osmosis
to win the competition for new seawater desalination capacity but it will allow
the RO process to replace distillation plants that are being retired in the Middle
East where the enormous capacity of these plants exists. The low pressure mem-
branes which are being developed for brackish applications will also further po-
tential municipal wastewater reclamation as an alternate source of potable and
industrial water. The reduced costs which will result from lower pressure/higher
rejection systems will effectively compete with pipelines, dams and other water
supply schemes that proliferate in the western United States.
While not totally essential to progress, it is suggested that oxidizing-agent-
resistant membranes will be developed in a thin film composite membrane.
This will broaden the applications available to RO and, at the same time, re-
duce the cost and complexity of existing plants which presently use chlorine
sensitive membranes.
An elusive goal has been the development of ion specific membranes. Based
on limited knowledge of worldwide research programs, it appears that this goal
will remain elusive for the forseeable future.
Finally, as the world becomes more aware of the environmental damage
caused by indiscriminate waste disposal it is apparent the RO process will play
a key role in mitigating that problem. It appears that the next market for the RO
process will be in industrial waste treatment in the United States to be followed
by application in other countries. Eventually, the world will be forced to use
reverse osmosis to reclaim municipal wastewater on a large scale and to put the
reclaimed water to a number of already demonstrated beneficial uses.
REFERENCES
1. El-Ramly, N.A. and Congdon. C.F., Desalting Plants Inventory, Report No.
6, United States Department of Interior (1977).
2. El-Ramly, N.A. and Congdon. CF., Desalting Plants Inventory, Report No.
7, National Water Supply Improvement Association (1981).
3. Wangnick, K., Desalting Plants inventory, Report No. 8, Water Supply Im-
provement Association (1984).
4. Wafer Desalination Report (M.C. Smith, Ed.), Vol. XXI, No. 37 (September
19, 1985).
5. Soltanieh, M. and Gill, W.N., Chemical Engineering Communications, Vol.
12, pp. 279-363.
6. Fluid Systems Division of UOP, Inc., Product Catalog (1984).
7. Beasley, J.K., “The Evaluation and Selection of Polymeric Materials for
Reverse Osmosis Membranes,” Proceedings of the International Con-
gress on Desalination and Water Reuse, International Desalination and
Environmental Association (1977).
8. Sudak, R.G. and McKee, M.E., “Low Pressure Composite Membranes,”
Proceedings of the International Congress on Desalination and Water
Reuse, International Desalination and Environmental Association (1977).
9. Orange County Water District, Evaluation of Membrane Processesand their
Role in Wastewater Reclamation, Vol. I (1979).
IO. Orange County Water District, Evaluation of Membrane Processesand their
Role in Wastewater Reclamation, Vol. I I (1980).
11. E.I. DuPont de Nemours & Co., Permasep Engineering Manual (1982).
12. Sourrajan, S., Editor, Reverse Osmosis and Synthetic Membranes, Nat-
ional Research Council of Canada (1977).
13. American Public Health Association, American Water Works Association
and Water Pollution Control Federation, Standard Methods for the
Examination of Water and Wastewater, 16th Edition (1985).
14. Saltonstall, C.W., Jr., Proceedings of the 10th Annual Conference of NWSIA
(1982).
15. Environmental Protection Agency, Environmental Regulations and Tech-
nology- The Electroplating Industry (1980).
16. Environmental Protection Agency, Economics of Wastewater Treatment
Alternatives for the Electroplating Industry (1979).
17. Vera, I., “Operating Experiences on a Reverse Osmosis Plant Which Con-
verts Seawater to Boiler Feedwater,” Industrial Water Engineering (No-
vember 1981).
5
Thin Film Composite Reverse Osmosis

Membranes
Robert J. Petersen and John E. Cadotte
A thin film composite reverse osmosis membrane can be defined as a multi-

layer membrane in which an ultrathin semipermeable membrane layer is de-
posited on a preformed, finely microporous support structure. This contrasts
with asymmetric reverse osmosis membranes in which both the barrier layer and
the porous substructure are formed in a single-step phase inversion process and
are integrally bonded.
Figure 5.1 contains a schematic diagram illustrating the concept of a thin
film composite reverse osmosis membrane. Alongside this is shown a photo-
micrograph of a fracture edge of an actual membrane of this type at 1,000 mag-
nification. In this photomicrograph, the barrier layer is not actually visible other
than as a smooth surface. Because its thickness is so small, it cannot be seen at
the magnification shown.
Fabrication of a thin film composite membrane is typically a more expen-
sive route to reverse osmosis membranes because it involves a two-step process
versus the one-step nature of the phase inversion film casting method. However,
it offers the possibility of each individual layer being tailor-made for maximum
performance. The semipermeable coating can be optimized for water flux and
solute rejection characteristics. The microporous sublayer can be optimized for
porosity, compression resistance and strength. Both layers can be optimized for
chemical resistance. In nearly all thin film composite reverse osmosis mem-
branes, the chemical composition of the surface barrier layer is radically dif-
ferent from the chemical composition of the microporous sublayer. This is a
common result of the thin film composite approach.
The term “thin film composite” has the connotation that the barrier layer
is extremely thin, and hence quite fragile. Indeed, the barrier layer may be quite
thin, varying to as low as 200 angstroms depending on the nature of the particu-
lar reverse osmosis membrane and its method of manufacture. But this does not
necessarily result in fragility. Some of these membranes may be considerably
307
Ultrathin Barrier
0.2pm
b)
Figure 5.1: (a) Schematic diagram of the cross section of a composite reverse
osmosis membrane; (b) scanning electron microscope photograph of the cross
section of a composite reverse osmosis membrane.
Thin Film Composite Reverse Osmosis Membranes 309
more rugged and chemically resistant than the typical asymmetric cellulose
acetate membrane in field service. It should be noted that the barrier layer in
asymmetric cellulose acetate membranes is itself only about 2000 angstroms
thick. Therefore, it may be more correct to refer to such membranes simply as
“composite” reverse osmosis membranes.
There are several potential routes to the preparation of composite reverse
osmosis membranes, whereby the ultrathin semipermeable film is formed or
deposited on the microporous sublayer.lr2 The film can be formed elsewhere,
then laminated to the microporous support, as was done in the earliest work on
this membrane approach. Or it can be formed in place by plasma polymerization
techniques. Alternatively, membrane polymer solution or polymer-forming re-
actants can be applied in a dipcoating process, then dried or cured in place. The
most attractive approach from a commercial standpoint, however, has been the
formation of the semipermeable membrane layer in situ by a classic “non-
stirred” interfacial reaction method. Several examples of membranes made by
this last approach have reached commercial status.
CELLULOSE ACETATE MEMBRANES
The first composite reverse osmosis membrane to be developed and described

consisted of an ultrathin film of secondary cellulose acetate deposited onto a
porous Loeb-Sourirajan membrane.3 The ultrathin film of cellulose acetate was
fabricated by a water surface float-casting technique. This has been described
to some extent in the published technical literature,4r5 and in considerable de-
tail in several reports on government-funded research projects.3j6” Figure 5.2
illustrates this process schematically.
In float-casting, a polymer such as cellulose acetate is dissolved in an aque-
ous solvent, usually cyclohexanone, to a level of about 5% by weight. A solvent
is preferred which has a slight solubility in water and a specific gravity of less
than 1.0 g/cc. When a casting dope of this type is allowed to flow down an in-
clined plane onto a quiet water surface, it shows a pronounced tendency to
spread over the water surface. Migration of the solvent into the water surface
(as well as some loss by air evaporation) occurs, leaving behind a floating solidi-
fied polymer film. Continuous addition of the casting dope with mechanical
drawing off of the solidified film, can be used to control the thickness of the
ultrathin film. For reverse osmosis membranes, thicknesses of 200 to 5000
angstroms have been achieved. In a related application involving float-cast thin
cellulosic membranes for hemodialysis, thicknesses of up to 2.5 @rn (25000
angstroms) were developed and used.’
The free floating film is transferred to a microporous support by bringing a
sheet of the support into contact with the underside of the ultrathin film, lifting
it from the water surface.” Also, two layers can be simultaneously cast and lam-
inated to a carrier web.‘,’ Numerous patents have appeared in recent years on
the fabrication of gas separation membranes by float-casting, even including
the double layer membrane technique.“r”
Ultrathin float-cast films exhibit visible light interference colors. These can
be used as a general guide for thickness, blue being about 2000 angstroms in the
case of cellulose acetate, gold being thinner, green and red being thicker. Actual
Figure 5.2: Schematic diagram of the float-casting of ultrathin cellulose acetate

membranes.
measurements have been made by interferometric methods on films deposited

and air-dried on glass plates.12r13
An alternate route to ultrathin cellulose acetate membranes exists via the
Carnell-Cassidy technique.‘4r’5 4n this method, a glass plate is mechanically with-
drawn at a slow, careful rate from a dilute solution of the polymer. A thin, air-
dried film is obtained, which could be released from the glass plate by immersion
in water. The result can be achieved on a large scale by the meniscus coating ap-
proach, illustrated in Figure 5.3.16 A commercial adaptation of this process was
developed by Lonsdale and Riley.“r’s .A microporous sheet of a cellulose ace-
tate/cellulose nitrate blend was first coated with a thin film of polyacrylic acid.
This coating was temporary in nature, intended to protect the microporous sheet
from solvent attack during overcoating with the semipermeable barrier film. The
polyacrylic acid would dissolve in water and wash out during subsequent usage
under reverse osmosis conditions. An ultrathin coating of cellulose triacetate
dissolved in chloroform was meniscus-coated onto the polyacrylic acid surface,
and dried. These membranes exhibited salt rejections of as high as 99%.
MlCAOPOROuS
SUBSTRATE
f
DIP
ROLLER
POLYMER
SOLUTION
Figure 5.3: Schematic diagram illustrating the meniscus coating technique.
During the period of 1965 to 1972, the best data on flux and salt rejection
for cellulose acetate membranes were exhibited by the composite membranes.
However, these membranes never reached commercial viability; efforts on them
died out completely by 1975. Reasons for this appear to be threefold. First,
composite cellulose acetate membranes were technically difficult to scale up.
Second, the advent of noncellulosic composite membranes in 1972 (the NS-100
membrane) offered much more promise for high performance (salt rejection and
water flux), especially for seawater desalination. Third, continual improvements
in asymmetric cellulose acetate membrane casting technology (such as the de-
velopment of swelling agents and of blend membranes) brought the performance
of asymmetric membranes to full equality with composite cellulose acetate
membranes.
MICROPOROUS POLYSULFONE SUPPORTS FOR COMPOSITE MEM-

BRANES
Early examples of cellulose acetate composite membranes used cellulose es-

ter sheet materials as the porous underpinnings for the float-cast films. These
sheet materials included: (a) Loeb-Sourirajan asymmetric cellulose acetate mem-
branes formulated to have loose, high flux structures and (b) mixed cellulose
ester microfiltration membranes in the tightest grades. These were concluded
to be far from optimum because of their susceptibility to compaction at the
seawater test condition of 1,500 psi prevalent at that time.
In the fall of 1966, researchers at North Star Research Institute began a
search for compression-resistant microporous substrates.‘” This effort resulted
in the development of microporous sheets of polycarbonate (Lexan) and poly-
sulfone (Udel).20 Figure 5.4 shows a graph comparing the flux levels and flux
stability for three membranes made at that time: (a) float-cast cellulose acetate
on microporous polysulfone, (b) float-cast cellulose acetate on a mixed cellu-
lose ester microfilter support and (c) a standard asymmetric cellulose acetate
membrane. The improvement in membrane fluxes was readily apparent, when
switching from cellulosic substrates to the microporous polysulfone substrate.
The structure of a microporous polysulfone sheet, optimized for composite
reverse osmosis membranes, is illustrated in Figure 5.5. These pictures were ob-
tained using freeze fractured samples in a Hitachi Model H-600 STEM electron
microscope. The SEM phor:igraphs show an asymmetric cross section of graded
porosity-very dense at the top surface and highly porous at the bottom surface.
High magnification views of the dense top layer show the anticipated nodular
structure, which has also been observed in asymmetric aromatic polyamide and
cellulose acetate membranes prepared by phase inversion methods. Surface
views of the microporous polysulfone indicate pores of 200 to 300 angstroms.
Polysulfone was recognized as a major improvement in the state-of-the-art
of composite membranes at that time. But the broad scope of its usefulness was
never fully appreciated until later. It has since become widely useful in its own
right as an ultrafiltration membrane. It was subsequently included in a U.S.
patent on ultrafiltration membranes by Michaels that issued in 1971.21 Con-
cerning reverse osmosis membranes, it represented a key development that later
enabled rapid progress to take place in noncellulosic composite membranes.
I I 1 I
membrane thickness 2000 R

annealing temperature 80°C
20 - acetyl content 39.4x
test conditions 3.5% NaCl, 1500 psi
15 -
polysulfone support--composite CA
10
Gelman GA-10 support--composite CA ’

n 0
CA anisotropic membrane
5&_
0
8 I I I I
OO 5 10 15 20 25
Time (hours)
Figure 5.4: Effect of support films on the reverse osmosis water flux of cellu-
lose acetate membranes.
Figure 5.5: Cross section and surface of a microporous polysulfone sheet used
in composite reverse osmosis membranes: (a) total cross section of a polysulfone
sheet cast on a nonwoven polyester fabric, then delaminated prior to freeze-
fracture for SEM (note fiber tracks on backside of the sheet); (b) backside of
sheet showing cellular structure, which extends through 85% of the sheet thick-
ness; (c) transition region from cellular to nodular structure near film surface;
(d) dense nodular structure at the surface; (e) high magnification of the extreme
top surface cross section; (f) high magnification view of the surface structure
showing the texture of the top surface.
NS-100 COMPOSITE MEMBRANE
The NS-100 membrane (initially designated as NS-1) was the first noncellu-
losic composite membrane to appear in the published literature and have an im-
pact on the reverse osmosis scene.22r23.This membrane, invented by Cadotte,%
consisted of a microporous polysulfone sheet coated with polyethylenimine,
then interfacially reacted with either 2,4-toluenediisocyanate (TDI) or with
isophthaloyl chloride (IPC). In the first case, a polyurea is formed; in the sec-
ond case, a polyamide. The chemistry of this membrane is as follows:
-CCH2CH2~-CH2CH2NH---
CH2CH2NH2
--CH2CH2r;J--cH2CH2tj- -CH2CH2N--CH2CH2~-
c=o tH2 c=o
CH2
A=o 0 =o
kH
E2 cNHH2
4
b=o
0 =o
AH
c=o
CH
(
4
Strictly speaking, the descriptors NS-1 and NS-100 (North Star Research In-
stitute) in the technical literature refer to the TDI-based polyurea membrane.
MSI-400 (Membrane Systems, Inc.) refers to the same composition.25 PA-100
(UOP, Fluid Systems Division) and NS-101 have been used to name the corre-
sponding IPC-based polyamide. The polyamide analog exhibits somewhat higher
flux and slightly lower salt rejection than the polyurea form of the membrane.
A low pressure version of PA-lOO/NS-101 membrane, optimized for brackish
water operation on Mohawk-Wellton (Yuma) agricultural drainage water, was
developed by UOP under the name TFC-202.26
Polyethylenimine is a product of the self-condensation of the strained-ring
compound ethylenimine (aziridine, azacyclopropane). It is a globular molecule
having a not quite statistical distribution of amine groups (30% primary, 40%
secondary and 30% tertiary amine). The most effective molecular weight range
for this water-soluble polymer in composite membrane fabrication has been
10,000 to 60,000.
Figure 5.6 illustrates the preparation of the NS-100 membrane. A micro-
porous polysulfone sheet is saturated with a water solution of the polymeric
amine (0.5 to 1.0% solution of polyethylenimine). Excess solution is drained off
TDI
HEXANE
surface of aqueous ultrathin heat-cured

polysulfone PEI polyurea PEI gel
substrate coating barrier
layer
Figure 5.6: Steps in the preparation of the NS-100 composite membrane.
by positioning the film vertically. The amine-impregnated film is then immersed

in a solution of 0.1% tolylene diisocyanate in hexane for 30 to 60 seconds. This
forms a very thin crosslinked polyurea zone on the surface of the wet polyethyl-
enimine layer. The film is then heated in an oven at 110°C for up to 15 minutes.
This treatment not only dries and anneals the film, but performs the crucial step
of crosslinking the residual unreacted polyethylenimine. Internal crosslinking of
the polyethylenimine takes place via ammonia elimination from adjacent amino
groups. If the polyethylenimine were not insolubilized such as by this treatment,
it would wash out later during use of the membrane. The polyurea layer itself
is too thin to withstand high pressures without the help of the polyamine gel
sublayer as an intermediate level support.
The NS-100 membrane is capable of giving salt rejections in excess of 99%
in tests on salt solutions simulating seawater (18 gfd, 3.5% synthetic seawater,
1,500 psi, 25°C). If the polyurea interfacial reaction step is omitted, and the
polyethylenimine-coated polysulfone film is heat-cured as usual, a crosslinked
polyethylenimine semipermeable barrier film is generated. This membrane gives
70% salt rejection and 55 gfd water flux under the same test conditions as above.
Also, if the fully formed NS-100 membrane is dried at 75’C, which is too low a
temperature to effectively crosslink the amine layer, the resulting film will ex-
hibit a salt rejection of 96% or less.
This membrane is highly sensitive to chemical attack by hypochlorite ion
and hypochlorous acid in chlorinated feedwaters. Wrasidlo prepared a variation
of the NS-100 membrane in which the primary amine groups of polyethylen-
imine were cyanoethylated before reaction with isophthaloyl chloride.27 This
membrane was claimed to be chlorine-resistant, but was probably not so.
Optimization studies on the NS-100 membrane were carried out by other
groups in addition to North Star Research Institute. Fang and Chian used a
statistically designed set of 33 experiments to produce a membrane with
10.30+ 1.95 gfd and 99.30+0.18% rejection tested on 5,000 ppm sodium chlo-
ride at 600 psi.*s Sudak and coworkers at Membrane Systems, Inc., also used
statistically designed experiments to develop a membrane (MSI-400) capable of
low pressure operation.*’ This membrane demonstrated a flux of 20 gfd and a
salt rejection of 97% when operated at 250 psi on a 5,000 ppm sodium chloride
feedwater. Modification of fabrication conditions produced a seawater version of
MSI-400 capable of generating 17 gfd and 99% salt rejection on 35,000 ppm so-
dium chloride at 800 psi and 24°C.
Chian and coworkers have also published studies on the organic solute rejec-
tions of the NS-100 membrane*’ and its potential for pesticide removal from
water.30
PA-300 AND RC-100 MEMBRANES
The NS-100 membrane, being the first of a kind, was by no means optimum
in chemistry and performance. Flux was only marginally attractive for desalina-
tion, sensitivity to chlorine was extreme, and the barrier surface was thin and
brittle. In commercial fabrication trials, overcoating of the membrane surface
with a layer of water-soluble polyvinyl alcohol was practiced in order to over-
come its brittleness and susceptibility to abrasion damage during handling and
spiral element fabrication.
A broad-based effort in both the United States and Japan has since taken
place to find other polymeric amine reactants that would contribute to better
properties. Perhaps the most significant outcome of this effort to date has been
the discovery of composite interfacial membranes based on “polyepiamine” by
Wrasidlo,31 and their development into the PA-300 and RC-100 commercial
forms by Riley and coworkers.32r33
Polyepiamine, also called polyetheramine, is the reaction product of poly-
epichlorohydrin with an excess of ethylenediamine. Its idealized structure and
its reaction products with isophthaloyl chloride (PA-3001 and 2,4toluenediiso-
cyanate (RC-100) are shown below:
H2N CH2CH2NH2
-CH2-CH-C- c
k “2
&I
-CH2CH-O- -CH2CH-O-
C”2 0 CH2 R
AHCH$H2NH&!NH NHCH2CH2NHC
I I
Reaction would take place on both the primary and secondary nitrogen
groups, of course. The end result would be a highly crosslinked structure, yet
retaining some ductility because of the flexible polyether backbone chains.
Initial data on the PA-300 analog were excellent.32 In tests on 35,000 ppm
sodium chloride at 1,000 psi, fluxes of 20 to 25 gfd and salt rejections in excess
of 99.4% (at pH 5.0 to 6.0) were observed. Similarly, in a brackish water test on
5,530 ppm sodium chloride at 400 psi, a water flux of 20 gfd at 98% salt rejec-
tion was observed. These data represented major decreases in operating pressure
while maintaining effective permeate production rates.
In other respects, it also shared the favorable characteristics of the NS-100
membrane. That is, it was resistant to pH 3 to 12, and showed far better com-
paction resistance than cellulose acetate. Also, it possessed the capability to op-
erate at elevated temperatures, though some irreversible flux decline could still
occur.34 Rejections of various organics were also good, as shown in Table 5.1
These were in sharp contrast to organic rejection data on cellulose acetate mem-
branes. Initially, PA-300 was also postulated to possess good chlorine resis-
tance.3’ Subsequent experience showed it to be equally sensitive to chlorine as
NS-100.
Table 5.1: Reverse Osmosis Performance of the PA-300 Composite Membrane

Toward Various Organic Solutes
Solute Concentration (ppm) l!! Solute Rejection (%)*
acetaidehyde 600 5.0 70-75

acetic acid 190 3.8 65-70
acetonitrile 425 6.3 25
Alcozyme (soap) 2,000 9.3 99.3
aspartic acid 1,500 3.2 98.3
2-butanone 465 5.2 94
butyl benzoate 220 5.8 99.3
citric acid 10,000 2.6 99.9
2,4-dichlorophenoxy-
acetic acid 130 3.3 98.5
dimethyl phthalate 37 6. 2 95
ethanol 700 4.7 90
ethyl acetate 366 6.0 95.3
glycine 1,400 5.6 99.7
phenol 100 4.9 93
phenol 100 12.0 99
tetrachloroethylene 104 5.9 93
trichlorobenzene 100 6.2 99
* Conditions: 1000 psi, 25’C
Both PA-300 and NS-100 exhibit mild cationic behavior because of the ex-
cess unreacted amino groups present in the barrier layer. One manner in which
this is evident is their propensity to absorb anionic surfactants and lose flux
thereby.% Consequently, wastewaters containing anionic surfactants cannot be

economically treated with this particular type of composite membrane.
Some difficulty was encountered in reliably producing this membrane dur-
ing its earlier history. A major problem was due to the nature and supply of
polyepiamine itself. It was originally developed by Dow Chemical Company as a
cationic coagulant for water treatment purposes. Membrane manufacture ac-
counted for only a tiny fraction of its usage. The polyepiamine polymer was ob-
served to be unstable in shelf storage, undergoing a continuous change in viscos-
ity, and sometimes turning into insoluble gel. Eventually, Dow Chemical ceased
production of this chemical, and membrane manufacturers desiring to use it
were forced to develop their own preparative methods for the polyepiamine.
The PA-300 membrane was the first composite reverse osmosis membrane
to be used successfully in a major seawater desalination facility-the 3.2 MGD
plant at Jeddah, Saudi Arabia.3s Much of the original membrane in this plant,
as well as replacement membrane, is believed to be RC-100 (the TDI-based ana-
log) because of its greater stability and retention of salt rejection. The Jeddah
plant was truly a pioneer installation for spiral-wound composite membranes.
Even though it was attended by a variety of start-up and operating problems,
mostly unrelated to the membranes, 36 it continues to operate successfully at
this date.
A low pressure version of PA-300, designated LP-300 membrane, has also
been developed for brackish water applications.26 This membrane provided up
to 24 gfd and 98.5% salt rejection operating on a simulated Yuma feedwater at
200 psi. A summary of UOP developments in the area of these various composite
membranes as of the end of 1980 has been published by Riley and coworkers.37
OTHER INTERFACIAL MEMBRANES BASED ON POLYMERIC AMINES
Various polyamines have been synthesized and evaluated in the fabrication

of the NS-100 type of membrane. These various compositions are described in
the patent literature. Some of these efforts have involved polymeric amines con-
taining only secondary amino groups to reach a goal of improved chlorine resis-
tance. Whether any of them have reached commercial status cannot be deter-
mined because of the current trend to avoid publication of the compositions
of new commercial reverse osmosis membranes.
Kurihara and coworkers at Toray Industries prepared several aminated deriv-
atives of polyepichlorohydrin, then formed composite polyamide membranes by
interfacial reaction with isophthaloyl chloride.38 Polyepichlorohydrin was con-
verted to polyepiiodohydrin, then reacted with either 4-(aminomethyl)piperidine,
3-(methylamino)hexahydroazepine, or 3-(amino)hexahydroazepine. Also, poly
epiaminohydrin was prepared by reduction of the azide derivative of poly-
epiiodohydrin. Best salt rejections were obtained if the polymeric amine for-
mulation contained a substantial proportion of the monomeric amines as co-
reactants in the interfacial reaction. In tests on 3.5% sodium chloride at 800 psi
and 25OC, salt rejections of 99.5% at fluxes of 8 to 9 gfd were characteristic. A
three-zone barrier layer was produced, consisting of a heat-crosslinked polya-
mine gel (as in NS-1001, a polyamide layer incorporating both the polymeric
and monomeric amine reactants, and an additional surface polyamide layer

comprised almost solely of the monomeric amine reactant combined with the
acyl halide. Toray has recently announced the commercial introduction of a
new, high rejection polyamide composite membrane, which may or may not
correspond to this patented composition.
Kawaguchi and coworkers at Teijin have prepared a series of polymers
based on poly(diallyl amine), its copolymer with sulfur dioxide, and various
terpolymers.39 The chemistry of this polymer synthesis is shown below. The
patent description shows the diallylamine polymers to be polypiperidine (six-
membered ring) derivatives, but there are a number of publications that show
this monomer to produce preferably polypyrrolidine (five-membered ring)
structures:
Cti2=yi ykcH2
CJJ2,CH2
- p”2J$H2!e
Y
H k
(ammonium
salt form) \’ so2 - tCH2~cH2*02~n
These polymers were then interfacially reacted with di- and trifunctional aro-
matic acyl halides to give polyamides. Because the only reactive amine groups in
these polymers were of the secondary amine type, such polymers should be
chlorine-resistant. Dynamic chlorine tests (5 ppm, pH 6.0 to 6.5, 0.5% NaCI,
600 psi, 25°C) of 40 to 80 hours duration appeared to uphold this inference.
Interestingly, the best salt rejections were observed for the polyurea analogs,
formed by crosslinking the polymers with isocyanates instead of acyl halides.
For instance, poly(diallyl amine) reacted with 2,4-toluenediisocyanate gave
membranes with 96.9 and 99.7% salt rejection (50 and 20 gfd flux respectively)
under the above test conditions. Salt rejections for the polyamide examples
rarely exceeded 95% in the patent examples. The polyurea analogs would not
be resistant to chlorine.
Several other systems of polymeric amine-based interfacial polyamide mem-
branes have appeared in patents by Kawaguchi and coworkers. One patent de-
scribes the use of amine-terminated oligomers prepared by reaction of polyepox-
ides with polyfunctional amines. Another patent describes the attachment of
polyfunctional amines as side groups onto linear soluble polymers via carbox-
amide or sulfonamide linkages.41 A third patent covers the use of additives to
the polymeric amine phase, these additives bringing about additional crosslink-
ing of the residual amino groups through heat-curing after interfacial formation
of the barrier layer.42 Examples of such additives include esters, chiorohydrins,

imidazoamides and carbamates. Finally, one patent describes the preparation of
amphoteric polyamide barrier layers containing both free carboxylate groups
and ammonium groups.43 Such membranes show high rejection levels towards
sucrose (92 to 99%) while freely passing sodium chloride (15 to 25% rejection).
Yaginuma patented interfacial membranes made by condensation of poly-
alicyclic diisocyanates and diacyl halides with poiyethylenimine or polyepia-
mine. 44 This approach was claimed to provide high organic rejections simulta-
neously with low salt rejections, whereas comparative data for typical aromatic
diisocyanates or diacyl halides showed high rejections for both types of solutes.
However, only a wastewater product, naphthalenesulfonic acid/formaldehyde
condensate, was used in the testing of such membranes.
No commercially available membranes corresponding to any of the above
series of patents have as yet been reported, with the possible exception of the
Toray patent.
INTERFACIAL POLYMERIZATION WITH MONOMERIC AMINES: NS-300

MEMBRANE
The initial studies by Cadotte on interfacially formed composite polyamide

membranes indicated that monomeric amines behaved poorly in this membrane
fabrication approach. This is illustrated in the data listed in Table 5.2, taken
from the first public report on the NS-100 membrane.** Only the polymeric
amine polyethylenimine showed development of high rejection membranes at
that time. For several years, it was thought that polymeric amine was required
to achieve formation of a film that would span the pores in the surface of the
microporous polysulfone sheet and resist blowout under pressure tHowever, in
1976, Cadotte and coworkers reported that a monomeric amin; piperazine,
could be interfacially reacted with isophthaloyl chloride to give a polyamide
barrier layer with salt rejections of 90 to 98% in simulated seawater tests at
1,500 psi.45 This improved membrane formation was achieved through opti-
mization of the interfacial reaction conditions (reactant concentrations, acid
acceptors, surfactants). Improved technique after several years of experience in
interfacial membrane formation was probably also a factor.
A typical formula for membrane fabrication consisted of an aqueous phase
containing 1 :I :0.5 weight percent piperazine:sodium hydroxide:dodecyl sodium
sulfate, and a hexane phase containing 1 .O percent weight/volume of isophthaloyl
chloride. This membrane exhibited up to 26 gfd and 98% seawater rejection
(3.5% synthetic seawater, 1,500 psi, 25°C) and up to 4 gfd and 99.2% magne-
sium sulfate rejection (0.5% magnesium sulfate, 200 psi, 25°C). Unfortunately,
seawater salt rejections in excess of 96% could not be produced routinely, and
brackish water fluxes were too low to be attractive.
To increase the flux of this membrane, partial or complete substitution of
isophthaloyl chloride with trimesoyl chloride was examined.46r47- Dramatic
changes in membrane flux and salt rejection were observed. Table 5.3 lists the
results of this approach. As the trimesoyl chloride content of the acyl halide
reactant was increased from 0 to 100%. seawater salt rejection dropped while
Table 5.2: Reverse Osmosis Performance of Membranes from Various Diamines

or Polyamines Reacted with Terephthaloyl Chloride
Reverse Osmosis Performance*
Water Flux Salt Rejection
Polyamine (gfd) (percent 1
hexanediamine 104 a.2

1,3-diaminopropane 233 30.6
ethylenediamine Too brittle to test
hydrazine Too brittle to test
p-phenylenediamine 53 38
m-phenylenediamine 26.9 12
piperazine 59.8 74
polyethylenimine 29.3 96.3
* Test Conditions: 3.5% synthetic seawater, 1500 psi, 25’C
Table 5.3: Seawater and Brackish Water Test Data on Membranes of Poly-
(piperazineamides)
Seawater Test** Brackish Water Test***
Acid Chloride Flux Rejection Flux Rejection
IPC TMC Acid Acceptor* (gfd) (I) (gfd) (4)
1.0 0.0 1.0% NaOH 26 98 4 99.2

0.9 0.1 0.9% Na3P04 33 96 If3 99.0
0.8 0.2 0.9% Na3P04 73 78 58 99.9
0.67 0.33 0.9% Na3P04 94 65 77 99.6
0.5 1.5 0.9% Na3P04 96 64 31 99.9
0.0 1.0 1% DMP a0 68 26 99.3
* Key: NaOH=Sodium hydroxide, Na3POq=trisodium phosphate, DMP= N,N’-

dimethylpiperazine
** Test Conditions: 3.5% synthetic seawater, 1500 psi, 25°C
*** Test Conditions: 0.2% magnesium sulfate, 200 psi, 25°C
magnesium sulfate rejection remained constant. Furthermore, flux rose to a

peak value at a 2:l isophthaloyl:trimesoyl ratio of acyl halides, then dropped
off. Note that the acid acceptor was changed in this series from sodium hydrox-
ide to sodium phosphate to N,N’-dimethylpiperazine. These changes were made
to achieve the highest combination of flux and magnesium sulfate rejection at
each acyl chloride combination.
The combination of piperazine with trimesoyl chloride in composite mem-
brane form was named NS-300. Trimesoyl chloride leads to a crosslinked poly
amide structure. Apparently, however, considerable formation of hydrolyzed
carboxylate groups also occurs. This is evidenced by the anion selectivity of the
membrane, demonstrated by the sequential salt rejection data in Table 5.4 for
a series of salts on test with a single set of membrane specimens.
Table 5.4: Effect of Cation and Anion Valence on Rejection of Various Salts by
a Piperazine Trimesamide Interfacial Polyamide Membrane
Reverse Osmosis Test Data*
Solute Used in Salt Rejection Flux
Test Loop (XI (gfd)
0.1% MgS04 98.0 35

0.1% NaCl 70 31
0.5% NaCl 50 42
0.5% Na2S04 97.8 41
0.5% M&l2 46 32
0.5% M&SO4 97.9 32
* Test conditions: 200 psi, 25°C
The Donnan ion effect is also reflected in the lower rejection of sodium
chloride at the higher salinity level (0.5% vs. 0.1%). The chemistry of this mem-
brane is shown as follows:
Synthesis of piperazine-terminated oligomers as prepolymers for interfacial

membrane formation was also examined. 46r48Excess piperazine was reacted with
di- and triacyl chlorides in an inert solvent such as 1,2-dichloroethane. The re-
sulting amine-terminated polyamide oligomers had low solubility in the solvent
system and precipitated. This served to limit the degree of polymerization of the
oligomer. Even so, a portion of the product was insoluble in water and was fil-
tered out during preparation of the aqueous oligomeric amine solution for the
interfacial reaction step.
Table 5.5 lists the best performance data obtained for some piperazine oligo-
mer membranes interfacially reacted with isophthaloyl chloride. The objective of
these tests was to achieve single-pass seawater desalination membranes. As such,
the presence of free carboxylate groups was avoided. Use was made of the tri-
mesoyl chloride or alternate triacyl halides in the oligomer formation step, and
diacyl chlorides in the interfacial reaction step. A few examples of seawater de-
salination membranes were obtained. Best results were seen for piperazine-
cyanurate prepolymers interfacially crosslinked by isophthaloyl chloride, but
fluxes were low in view of the operating test pressure of 1,500 psi.
Table 5.5: Reverse Osmosis Properties of Interfacial Membranes Formed of

Piperazine Oligomers and lsophthaloyl Chloride
Reverse Osmosis Test Data*
Reactant used to Prepare Acid Flux Rejection
Piperazine Oligomer Acceptor (gfd) (%)
trimesoyl chloride NaOH 13 99.0
trimesoyl chloride triethylamine 58 93.0
cyanuric chloride NaOH 14 99.2
cyanuric chloride triethylamine 24 98.0
phosphorus oxychloride N, N’-dimethyl- 45 93.9

piperazine
1:l trimesoyl chloride: triethylamine 34 92.4

isophthaloyl chloride
* Test conditions: 3.5% synthetic seawater, 1500 psi, 25’C
The use of piperazine-terminated oligomers in composite membrane forma-

tion has been explored by several working groups. Even high polymers contain-
ing piperazine pendant groups have been made. The interest in piperazine stems
in part from the work by Credali and Parrini on asymmetric poly(piperazine-
amide) membranes, wherein this type of polyamide membrane is reported to be
resistant to chlorine in chlorinated feedwaters.49r50.It is of some surprise to note
that an interfacial polyamide membrane formed from piperazine and acyl hal-
ides is not fully chlorine resistant. ” In a long term supervised test on brackish
water at the Roswell Test Facility, some chlorine damage was beginning to ap-
pear within 2,000 hours continuous exposure to 0.75 ppm active chlorine at
pH 5.5, as shown in Figure 5.7.
NF-40 COMPOSITE MEMBRANE
A reverse osmosis membrane is commercially available, named NF-40

(FilmTec Corporation), which is closely based on the NS-300 membrane tech-
nology. Typical solute rejection data for this membrane are as follows: sodium
chloride, 45%; calcium chloride, 70%; magnesium sulfate, 93%; sucrose, 98%.
Water flux of the membrane averages about 23 gfd at 225 psi and 25°C. As al-
ready noted for NS-300, the sulfate anion is associated with high rejection; the
chloride anion, low rejection. Partial discrimination between monovalent and di-
valent cations (sodium versus calcium) has also been observed for NF-40. The
membrane can be operated at temperatures to 45°C and in a pH range from 3
to 11. In this respect, it would find probable use in industrial separations where
700 _I i 70
600 -60
400
300
200
Figure 5.7: Long term test of NS-300 membrane on chlorinated brackish water
showing onset of membrane failure at 2,500 hours due to chlorine attack.
the temperature and pH limitations of “loose” cellulose acetate asymmetric

membranes are a problem.
An interesting behavior of this kind of membrane is its response to feed-
water salinity and operating pressure. This type of membrane will show changes
in flux and salt rejection as ionic strength of the feedwater is varied. This is il-
lustrated in Figures 5.8 and 5.9 for two types of salts, sodium chloride and mag-
nesium sulfate. The changes in membrane flux as a function of feedwater salin-
ity are not explained by osmotic pressure differences across the membrane.
Among other applications, this membrane is being used in the treatment
of salt whey, a waste stream product of cheese making in which the sodium
chloride content is very high (up to 7%). The salt whey protein and lactose can
be both concentrated and desalted in a single membrane operation (requires
some diafiltration).
NTR-7259 COMPOSITE MEMBRANE
A similar type of membrane, named NTR-7250, is being marketed by

Nitto Electric Industrial Company.” This membrane exhibits high rejection
of sodium and magnesium sulfates (95 to 98%) and low rejection of sodium
chloride. Compared to NF-40 membrane, its water flux and sulfate rejections
are both higher, and long term chlorine resistance is claimed. It has been op-
erated successfully for 4,500 hours on a chlorinated tapwater containing about
one part per million average chlorine content, as shown in Figure 5.10.
Potential applications exist in industrial separations, water softening, and
point-of-use filtration for high purity water in semiconductor chip manufactur-
NOUXiNY 1-M Y
Test Conditions: 150-190 pS/cm tapwater, 286 psi

120 12-20°C, pH 6.4-7.0, 0.7-1.2 ppm C12,
2-inch spiral element, 1.6 gpm brine flow.
Product
Conductivi .ty 8o
(vS/cm)
50
1 I I
C
200 -
Flux at
25’C
(gpd) go 0 00 0 0 008 o~ooo~o 0 000 000 0
100 -
, I I I
1000 2000 3000 A00 53GO
Elapsed Time (hours)
Figure 5.10: Long term performance of NTR-7250 during exposure to chlo-

rinated tapwater.
ing. It has one potential drawback in that pH resistance is limited to pH 9 on

the alkaline side. This limits its applications where alkaline feedwaters and alka-
line cleaning are involved.
Although its composition has not been revealed, it may consist of, or be re-
lated to, the composition described in a German patent application by Kamiyama
and coworkers.s3 This patent application describes a membrane made by inter-
facial reaction between a 1.0% hexane solution of trimesoyl chloride and an
aqueous solution containing 0.25:0.25:0.5% by weight polyvinyl alcohol:piper-
azine:sodium hydroxide. This membrane, after curing at 110°C for IO minutes,
exhibited 30 gfd water flux and a magnesium sulfate rejection of 98.9% at
200 psi on a 500 ppm salt solution. In a recent review by Ohya on current
progress in reverse osmosis membranes,s4 a (somewhat overly simplified) compo-
sition is given for the NTR-7250 membrane consisting of both ester and piper-
azineamide groups. This composition, shown below, is consistent with the patent
application and with the pH 9 upper limit for alkaline stability:
-CH$H-
C+o
CIOC 0 E-t@
0
FT-30 COMPOSITE MEMBRANE
Cadotte discovered that aromatic diamines, interfacially reacted with triacyl

halides, gave membranes with dramatically different reverse osmosis perform-
ance characteristics than membranes based on aliphatic diamines.ss~s6 Before
that time, the area of aromatic amines in interfacial membrane formation had
been neglected because of two factors: (a) the emphasis on chlorine-resistant
compositions, which favored use of secondary aliphatic amines such as piper-
azine, and (b) poor results that had been observed in early work on interfacial
aromatic polyamides. The extensive patent network in aromatic polyamide
(aramid) technology may also have been a limiting factor.
A typical recipe for an interfacially formed aromatic polyamide composite
membrane comprised a 2.0% aqueous solution of the aromatic diamine and a
0.1% nonaqueous solution of trimesoyl chloride. This recipe was extraordinarily
simple, and ran quite contrary to experience with piperazine-based membranes.
For example, surfactants and acid acceptors in the aromatic diamine solution
were generally not beneficial, and in many cases degraded membrane perform-
ance by lowering salt rejection. In contrast, surfactants and acid acceptors were
almost always beneficial in the NS-300 membrane system. In the nonaqueous
phase, use of isophthaloyl chloride as a partial replacement for trimesoyl chlo-
ride had relatively little effect on flux, but tended to decrease salt rejection and
increase susceptibility to chlorine attack.
Another feature of this type of membrane was the rough surface of the
membrane on a microscopic scale. As illustrated in Figure 5.11, interfacial aro-
matic polyamides from trimesoyl chloride and various aromatic diamines all
showed a well developed “ridge and valley” structure. For aliphatic polyamines
such as polyethylenimine, 1,6_hexanediamine, piperazine, and even &a’-dia-
minoxylene, this ridge-and-valley structure was absent; instead, a flat surface-
sometimes smooth, sometimes grainy-was observed. The thicker sections of this
aromatic polyamide membrane are apparently about as active in water permea-
tion as the thinner sections. The backside of the barrier layer, shown in Figure
5.12, contained numerous micropores or passageways for exit of permeate wa-
ter from the depths of the interfacial membrane.
Average thickness of the polyamide barrier layer is about 2000 angstroms.
Thus, this type of membrane has a barrier layer thickness approximately equiv-
alent to the skin thickness in asymmetric reverse osmosis membranes. The ridge-
and-valley structure does not appear to cause any increased susceptibility to
membrane fouling. Deposition of foulants to the point restricting flux and in-
creasing salt rejection has been found to be controlled by feedwater quality, ele-
ment design, and hydrodynamic parameters rather than the surface topography
of the membrane.s7
FT-30 membrane is made from one of the simplest aromatic diamines: 1,3-
benzenediamine. The final chemical structure of the membrane is believed to
be as follows:
CIOC
G2,,,
+0
0n coc’
COCI
- 2
Figure 5.11: SEM photographs of the surface texture of composite polyamide

membranes from aliphatic and aromatic amines: (a) uncoated microporous poly-
sulfone; (b) polyamide from polyethylenimine and trimesoyl chloride; (c) tri-
ethylenetetramine and trimesoyl chloride; (d) 1,3-benzenediamine and trimesoyl
chloride; (e) 2,4toluenediamine and trimesoyl chloride; (f) 4-methoxy-1,3-
benzenediamine and trimesoyl chloride. Note the smooth surface for aliphatic
amine-based interfacial trimesamides and the coarse ridge-and-valley structure
for aromatic amine-based interfacial trimesamides.
Figure 5.12: Topside and underside of the FT-30 composite reverse osmosis
membrane: (a) topside showing well-developed ridge-and-valley structure, and
also an area of membrane barrier layer folded over upon itself; (b) underside of
the barrier layer (foldover zone) showing the network of micropores inside the
ridge-and-valley structure.
Some hydrolysis of the trimesoyl chloride takes place during membrane fabrica-
tion. ESCA studies indicated that approximately one-sixth of the carboxyl
groups are present as ionic carboxylate and five-sixths of the carboxyl groups
are present as amides, leading to the above structure. The FT-30 barrier layer
is insoluble in sulfuric acid and in all organic solvents, in agreement with the
crosslinked nature indicated above. Its chemical structure is somewhat similar
to the composition of the duPont Permasep B-9 hollow fiber polyamide, be-
lieved to be approximately as follows:
Both membranes show similar behavior toward oxidants such as chlorine

and bromine,‘s and both membranes show approximately equivalent behavior in
organic rejections.s9 The crosslinking inherent in the FT-30 membrane barrier
layer appears to confer greater membrane stability towards compaction, halogen
oxidation and pH extremes. The FT-30 membrane also appears to have a uniquely
high rejection (95% or higher) of soluble reactive silica, which is advantageous in
ultrapure semiconductor water preparation. The high silica rejection may be a
function of the combination of the carboxylate ionic groups and the degree of
polymer crosslinking.
The properties of FT-30 membranes have been reviewed in several publica-

tions, including reverse osmosis performance under seawater and brackish water
test conditions.60”2 In commercially produced spiral-wound elements , the FT-
30 membrane typically gives 99.1 to 99.3% salt rejection at 24 gfd flux in sea-
water desalination at 800 psi and 25°C. In brackish water applications, FT-30
spiral elements can be operated at system pressures of as low as 225 psi while
producing water at 22 to 24 gfd. Similar flux levels are possible with the TFC-
202 and LP-300 membranes, as mentioned earlier. But it is notable that those
membranes achieve such high fluxes through use of extremely thin surface bar-
rier layers about only one-tenth the thickness of the FT-30 barrier layer.
Applications for FT-30 membrane have appeared in all reverse osmosis fields
from seawater desalination to home tapwater systems operating on line pressure.
At this date, it is the only commercial reverse osmosis membrane other than
cellulose acetate that has specific FDA approval for food contact usage.63 Ver-
sions of this membrane, manufactured under license to FilmTec, are available
in tubular form (ZF-99, Patterson Candy International) and plate-and-frame de-
sign (HR-95, HR-98, De Danske Sukkerfabrikker).64t65.
Chlorine resistance tests on FT-30 membranes, whether static jar storage
tests or dynamic tests with chlorine added to the feedwater, showed a much
lower rate of oxidation compared to other polyamide membranes such as the
NS-100. A peculiar property of the FT-30 membrane in regard to chlorine attack
was that the rate of oxidation was lowest in an acid pH range of 5 to 6 and higher
in the alkaline pH range.’ This is illustrated in Figure 5.13. This effect has also
been observed by Glater and coworkers.‘s The pH effect on chlorine oxidation
of FT-30 membrane was opposite of that normally observed with other mem-
branes, because the hypochlorous acid at pH 5 to 6 would have a higher oxida-
tion potential than sodium hypochlorite in the alkaline pH range. At pH 3, the
effect of hypochlorous acid on FT-30 membrane lifetime could not be measured
because the underlying polysulfone support layer was instead rapidly attacked
and weakened.
Chlorine attack on reverse osmosis membranes is believed to be catalyzed by
transition metal ions such as iron, manganese and cobalt. Figure 5.14 shows the
effect of ferric ion in promoting chlorine attack, using in this case FT.30 mem-
branes preimpregnated with ferric chloride. The presence or absence of these
heavy metal ions may explain the discrepancies in chlorine resistance that have
been reported for basically identical membranes such as NS-300 and NTR-7250.
The prepolymer approach which worked well for piperazineamide inter-
facial membranes was not useful in the case of FT-30 and its analogs. Poor solu-
bility of aromatic amide precursors in aqueous media was the main obstacle. A
patent application has appeared on the use of prepolymer formed from 1,3-
benzenediamine and trimellitic anhydride acid chloride.66 This prepolymer con-
tains a free carboxylate group and is soluble in water as the sodium salt. A mem-
brane is obtained by reaction of this intermediate with trimesoyl chloride, fol-
lowed by a curing step at 1 IO0 to 13O’C. Salt rejections of 98.5 to 99.1% on
2,000 ppm sodium chloride solution at 200 psi were obtained; fluxes were 4 to
11 gfd.
100
pH 5
90
60
70
60
50 P+ PH 8
40
Test Conditions:
30
3.5X synthetic seawater
\ 800 psi
20 pH 12 25’C
10
0 I I I I I I I 0 II I 1 11 11 11
3 5 7 9 11 13 15 17 19
EXPOSURE TIME (HOURS)
Figure 5.13: Effect of pH on the chlorine resistance of the FT-30 composite

membrane.
100 .m==---
I I I I I I
-__~~---------_-_-___
---_ -x
l ---_
\o
---_ - - -0
\a
\B
80 A
S? Xr 0.2% NaCI, 2OOPSi, 25’c

0: 0.2% MgCl2,260 Pai, 260NC
1
.-‘0 60
z
.L
E
f 40 \ F&l3 - impregnated membranes
\
0-J \
20
d
0
40 80 120 180 200 240 28IO
Operating time at 40 ppm Cl2 thoursl
Figure 5.14: Effect of ferric chloride treatment on the rate of chlorine damage
to FT-30 polyamide membrane.
NF-50 COMPOSITE MEMBRANE
A new variation related to the FT-30 membrane is being developed-the NF-

50 composite membrane-which would appear to occupy a unique place in mem-
brane technology. The NF-50 membrane has approximately the same character-
istics as NTR-7250 and NF-40, but possessesan extremely high water flux. Re-
verse osmosis operation in large systems at a pressure of 35 to 50 psi is possible.
The NF-50 membrane thus becomes the first example of a reverse osmosis mem-
brane capable of operation at ultrafiltration membrane pressures.
Characteristics of this membrane include 30 to 40% sodium chloride rejec-
tion, 85 to 90% magnesium sulfate rejection, 98% sucrose rejection, and 99%
raffinose rejection. Water flux is 15 to 25 gfd at 35 to 50 psi, depending on feed-
water composition. The effect of increasing salinity on salt rejection and water
flux is similar to the behavior observed for NF-40 membrane as was illustrated
in Figures 5.8 and 5.9.
Applications for NF-50 membrane potentially include industrial separations,
partial water softening, removal of trihalomethane precursors from groundwaters,
and point-of-use filtration of high purity water for semiconductor chip manufac-
turing.
MECHANISM OF INTERFACIAL MEMBRANE FORMATION
In the book, Condensation Polymers: By Interfacial and Solution Methods,

by P.W. Morgan, 67 interfacial polyamide formation is stated to occur in the or-
ganic phase, that is, on the organic solvent side of the interface. Several proofs
are presented in support of this statement. For instance, monofunctional acyl
halides added to the difunctional acyl halide in the organic phase always lowered
polymer molecular weights. However, monofunctional amines added to the di-
functional amines in the aqueous layer did not always show this effect. In this
latter case, partition coefficients became a factor, particularly when relative re-
activities of the amines were comparable. Mass transfer rate of diamine across
the interface into the organic phase was noted to be the rate-controlling step
at all concentrations of diamine.
A visually graphic display of this mechanism is illustrated in Figure 5.15. An
insoluble colored powder, deposited on the face of an interfacial polyamide film,
becomes incorporated into the polyamide film if the powder is introduced on
the organic phase side. However, if the powder is deposited on the aqueous
phase side of the growing interfacial film, it remains loose and unattached.
In the case of a polymeric amine such as polyethylenimine or polyepiamine,
the solubility of the polymer in the organic phase would be very low. Reaction
would take place at the interface to form an extremely thin, crosslinked net-
work. This network would block the transport of further polymeric amine from
the aqueous phase into the organic phase. Continued buildup of the membrane
material on the organic side becomes impossible. Thereafter, the growth in thick-
ness of the barrier layer is controlled by the much slower diffusion of acyl halide
or isocyanate into the aqueous phase. As a result, composite membranes made
by the NS-lOO/NS-101 type of approach will naturally tend to have very thin
barrier layers-typically 200 to 250 angstroms thick.
- - - - -Ic
sebacyl chloride in 1,6-hexanediamine

xyene in water
insoluble dye powder
* polyamide membrane
‘a.4 ,‘.,H,‘-
AL_ __--L -interface
1 min 20 hr
1,6_hexanediamine
in water sebacyl chloride in
carbon tetrachloride
Figure 5.15: Visual demonstration of the direction of growth of a polyamide

film at the solution interface.
In the case of monomeric amines such as piperazine or 1,3-benzenediamine,

transport of the amine across the water-solvent interface takes place readily. Fur-
thermore, the interfacially formed polymer film remains rather porous to salts
and small molecules (until dried or heat-cured), so that membrane material con-
tinues to form on the organic side. Therefore, barrier layer thicknesses as high as
2500 angstroms are readily produced.
Monomeric amines have two advantages over polymeric amines in interfacial
composite membrane fabrication. First, monomeric amines can be obtained in
most cases as pure crystalline compounds, identical in lot after lot. Polymeric
amines, on the other hand, will show variations in purity, molecular weight,
chain branching and viscosity from lot to lot. This adds an element of variability
to the membrane fabrication process. Second, monomeric amines lead to thicker
barrier layers, which consequently tend to show better abrasion resistance and
greater tolerance to chemical attack. By contrast, a membrane such as PA-300
is normally overcoated with a protective layer of water-soluble polyvinyl alcohol
to minimize abrasion and salt rejection lossesduring spiral element assembly.
In the patent by Kurihara, Uemura and Okada, combinations of a poly-
meric amine with a monomeric amine were used to produce composite poly-
amide membranes having high salt rejections. The membranes were described
as having a bilayer polyamide barrier film: a surface polyamide zone rich in
monomeric amine, and a subsurface polyamide zone incorporating both mono-
meric and polymeric amine. This patent disclosure demonstrated an under-
standing of the mechanism of interfacial polyamide barrier layer formation.
A patent has also appeared by Fukuchi and coworkers on the combination
of polyethylenimine and several monomeric amines (including piperazine, 2,5-
dimethylpiperazine, and 4-aminomethylpiperidine) in interfacial polyamide
membranes. This patent also claimed very high salt rejections.
SULFONATED POLYMER COMPOSITES: NS-200 MEMBRANE
Several composite membranes have been prepared based on sulfonated poly-

mers. These are typically formed on microporous polysulfone supports by solu-
tion-coating techniques. One of the earliest and most unique examples of this
type of membrane was the NS-200 membrane. This membrane was discovered
by Cadotte,69 and was characterized in government-sponsored research pro-
grams. 7or71145The NS-200 membrane in its optimized formulation71 was pre-
pared by dipping a microporous polysulfone sheet in a 2:2:1 solution of fur-
fury1 alcohol:sulfuric acid:Carbowax 20M (Union Carbide) polyethylene oxide
dissolved in 80:20 water:isopropanol. Excess coating solution was drained away,
followed by an oven curing step at 125’to 140°C. The high oven temperature, in
addition to drying the membrane, generated a black sulfonated polyfurane resin
that was embedded in the surface of the polysulfone.
When the NS-200 membrane was initially discovered,7o only furfuryl alcohol
and sulfuric acid in water were used. The furfuryl alcohol was fairly rapidly
converted to a resinous product in the aqueous solution; isopropanol was even-
tually added as a cosolvent to prolong the usable life of the dip-coating solution.
Various water-soluble additives were also evaluated for their effects on forma-
tion of the sulfonated polyfurane resin and its subsequent reverse osmosis per-
formance. Examples of additives included saccharides, polyalcohols, polycar-
boxylic acids and polyethylene glycols. Of these, the 20,000 molecular weight
polyethylene oxide was singularly beneficial in increasing membrane flux quite
significantly without disturbing salt rejection. It is probably severely degraded
by sulfuric acid during the oven treatment step, and may be mostly absent in
the final membrane after washout.
The most outstanding property of the NS-200 membrane was its extremely
high salt rejection. In simulated seawater tests, salt rejection levels of 99.8 to
99.9% were rountinely observed. Typical performance in a 1,500 psi test on
3.5% synthetic seawater at 25°C was 20 gfd and 99.9% salt rejection, In a few
instances, laboratory samples of NS-200 membranes with fluxes of 40 to 50
gfd under these same test conditions were obtained.
In subsequent efforts on the commercialization of this membrane, in flat
sheet form, water fluxes of up to 23 gfd (99.6% salt rejection) at 800 psi on
simulated seawater” and 50 gfd (98% salt rejection) on 0.5% sodium chloride at
250 psin were achieved.
The freshly prepared membrane is glossy black in appearance, turning
brown in the wet state, and partially decolorizing in exposure to the bleaching
effects of sunlight. The chemistry of the NS-200 barrier layer in its early stages
of formation is believed to be approximately as shown below. But the barrier
layer composition is believed to be ultimately quite complex due to ring open-
ing crosslinking and other side-reactions.
Membrane instability was believed to result from hydrolytic cleavage of sul-

fate and sulfonate esters in the sulfonated polyfurane resin. This hydrolysis
would entail some depolymerization of the resin and concurrent formation of
ionic sulfonate groups, causing the membrane to absorb water and swell in
monovalent salt solutions. Oxidation by dissolved oxygen is another potential
mechanism that was not fully considered during attempted development of the
NS-200 membrane, but which may have been an important factor (cf. PEC-1000
membrane). Further studies by other groups attempting to commercialize the
membrane were unsuccessful in solving the NS-200 membrane instability prob-
72-74
lem.
PEG1000 MEMBRANE
The PEC-1000 membrane, developed by Toray Industries, has many similar-

ities to the NS-200 membrane. This membrane has been described by Kurihara
and coworkers as a thin film composite membrane in which the barrier layer was
formed by an acid-catalyzed condensation reaction on the surface of micro-
porous polysulfone. ” The cross section of this membrane, under high magnifica-
tion in a scanning electron microscope is quite similar in structure to the NS-200
membrane. The barrier layer is 300 angstroms thick.
The composition of the PEC-1000 membrane has not been specifically pub-
lished. However, it is covered by U.S. Patent 4,366,062.76 This patent describes
membranes formed by acid-catalyzed condensation of the monomer 1,3,5-tris-
(hydroxyethyl)isocyanuric acid (THEIC). By itself or in combination with ad-
ditives such as sorbitol or polyethylene glycol, this monomer leads to mem-
branes with 95 to 97% sodium chloride rejection (0.25% NaCI, 570 psi, 25”C),
but at extremely low fluxes (0.5 to 1.0 gfd). Organic rejections were very good.
When furfuryl alcohol was added as a comonomer to the THEIC, water
fluxes were increased tenfold. In addition, the extremely high salt rejections
characteristic of NS-200 were obtained, while the high organic rejections charac-
teristic of the isocyanurate moiety were retained. A typical patent example of
membrane fabrication uses a water solution of 1:2:2: 1 weight percent THEIC:fur-
fury1 alcohol:sulfuric acid:dodecyl sodium sulfate, deposited on microporous
polysulfone and cured at 150°C for 15 minutes. This membrane, possessing a
thin active layer 100 to 300 angstroms thick, showed 99.9% rejection and 12
gfd flux under seawater test conditions at 1,000 psi.
The chemical structure of the PEC-1000 membrane is probably quite com-
plex. This membrane, in its early stages of formation, may involve some of the
following types of chemical structures, for example:
HOCH2CHz‘N A
ti~CH*CH20H + UC, OH H2Xb
&,~O 2
hH2CH20H
-I
_CH2C”2\N a p+C”2 -n.d0 Ct.+ 0 a2- (etc)
JY4~
++CH2CH2 .CH2CH2-
Y
To H2CH20-
Basic characteristics of the PEC-1000 membrane include very high salt re-
jections, high organic rejections, and maintenance of high salt rejections over a
pH range of 1 to 13 (in short term tests).77 The membrane is not chlorine-resis-
tant, and, indeed, has been found to suffer damage from dissolved oxygen in
the feedwater. This has been alleviated by using sodium bisulfite as a feed-
water additive, and by use of a vacuum deaerator as a pretreatment step in a
large seawater plant design.7a Flux of the PEC-1000 membrane is low relative
to polyamide composite membranes; but for organic chemical concentration and
seawater desalination applications, the high rejection characteristics offset this
disadvantage.
Figure 5.16, adapted from Kurihara, 79rs0.shows a comparison of several
types of commercial reverse osmosis membranes in terms of salt rejection and
permeate flow rate under seawater test conditions (35,000 ppm, BOO psi, 25OC).
This chart emphasizes the capability of PEC-1000 to provide complete single-
stage seawater desalting. In a test at Toray’s Ehime desalination test facility on
42,000 ppm seawater (equivalent to Red Sea salinity), PEC-1000 spiral ele-
ments operated at 35% recovery produced a permeate having an average salinity
of only 220 ppm, well below WHO standards. Average salt rejection was 99.5%.
The type of display in Figure 5.16 has been adapted and altered by others,
such as Kamiyama and coworkers8’ It should be noted that flux is shown on a
logarithmic scale in Figure 5.16, and this presents a visual bias to the reader. If
r f
3.57. seavater
800 pst, 2s”c
99.5
99.9
- Hollow fiber
l - membranes
Thin film camposice

membranes
______________
Partial double-
PT-30 membrane
stage process
98
97
_ membranes
.05 .l .2 .3 .5 1 2 3 5
Permeate Flow Rate (m’/m2* day)
Figure 5.16: Comparison of desalination performance of different commercially

available membranes for seawater.
the flux axis were changed to an arithmetic scale, the flux advantages of poly-
amide composite membranes would become visually more evident. In the case
of seawater desalination plants, two stage systems become economically feas-
ible at the higher fluxes. In tapwater and brackish water purification systems,
absolute flux is normally more important than small differences in rejection.
In terms of organic rejections, PEC-1000 membrane shows the highest val-
ues among all commercial reverse osmosis membranes. Table 5.6 lists rejection
data for a variety of organic compounds. In most cases, these were measured at
solute concentrations of 4 to 5%. which represents a severe test protocol. Or-
ganic solute rejections determined for other commercial membranes were typi-
Table 5.6: Reverse Osmosis Performance of PEC-1000 Membrane for Various

Organic Solutes in Aqueous Solutions
Operating
Performance* Conditions
Functional Molecular Rejection Flux Concentra-
Group Solute Weight (X1 (gfd) tion (X) pH
Alcohol Methanol 32 41 9.7 5 6.9

Ethanol 46 97 5.9 6 6.9
Isopropanol 60 99.5 a. 1 5 6.5
n-Butanol 74 99.4 6.6 4 7.0
Benzylalcohol 108 02 2.5 4 7.0
Ethylene glycol 62 94 9.2 5 6.8
Propylene glycol 76 99.4 3.1 10 7.0
Glycerine 92 99.8 9.7 5 7.0
Phenol 94 99 6.1 1 5.2
Carboxylic Acetic Acid 60 86 9.7 b 2.6

acid Propionic acid 74 98 7.9 10 3.4
Oxalic acid 90 99.1 18 0.5 1.8
Aldehyde Formaldehyde 30 66 5.3 b 5.3
Ketone Acetone 58 97 7.4 4 6.7

2-Butanone (MEK) 72 98 5.3 4 6.6
Ester Ethyl acetate a0 99.2 4.6 4 6.8

n-Butyl acetate 116 99.6 9.7 1 6.9
Ether Tetrahydrofuran 72 99.8 7.1 5 6. 7

1,4-Dioxane a0 99.9 10 5 6.6
Amine Ethylenediamine 60 99.5 2 5 12.1

Aniline 93 95 2. a 1 8.3
n-Butylamine 109 99.9 3.3 1 7.0
hydrochloride
Amide Urea 60 85 14 1 6.9

N,N-Dimethyl- 73 98 a. i 5 6.5
formamide
N, N-Dimethyl- a7 99.6 7.6 5 6.5
acetamide
E-Caprolactam 113 99.9 9.7 5 6.5
Sulfoxide Dimethylsulfoxide 78 99.6 8.6 5 6.4
* Test Conditions: 800 psi, 25”C, pH and concentrations as indicated.

tally at solute concentration levels of 0.1 to 1.0%. Figure 5.17 gives a capsule
illustration of the differences in organic rejections for four commercial mem-
branes: PEC-1000, FT-30, Permasep 8-9 polyamide and asymmetric cellulose
acetate. The superior organic rejection characteristics of PEC-1000 are quite
evident in this graph.
60
Solute
Rejection 50
(“/.I
PEC-1000 composite
FT-30 composite
Asymmetric aramid (hollow fiber)
Asymmetric cellulose acetate
Figure 5.17: Comparison of organic solute rejections of various membranes.
SULFONATED POLYSULFONE MEMBRANES
Sulfonation of aromatic polymers has been explored as a method to pro-

duce hydrophilic polymers with water permeability and salt rejection character-
istics. These have been of interest because of their potentially high degree of
chlorine resistance. The use of sulfonated aromatic polymers for reverse osmosis
membranes began in the late 1960’s with the work of Plummer, Kimura and
LaConti of General Electric Company. 82 Polyphenylene oxide [poly(2,6-di-
methylphenyleneoxy)] were cast from a chloroform/methanol solvent. The

degree of sulfonation was limited so that the product would be water-insoluble.
Later, composite membranes of this polymer were prepared by meniscus coating
onto Celgard microporous polypropylene film and eventually onto microporous
polysulfone sheeting.s3 In the latter case, a solvent blend was developed for the
sulfonated polyphenylene oxide, consisting of nitromethane/methanoI/butanol,
which did not dissolve the polysulfone substrate. While this membrane appeared
useful for several difficult applications involving highly acidic or high tempera-
ture feedwaters, its stability and performance characteristics were not sufficient
to result in its commercialization.
After the initial work on asymmetric sulfonated polyphenylene oxide mem-
branes, sulfonated polysulfone was subsequently examined by several research
groups. However, the appearance of patents in 1973 and 1977 by Rhone-
Poulenc~r8’. covering asymmetric membranes of sulfonated polysulfones ap-
peared to dampen such research efforts. Also, a basic problem with sulfonated
polysulfone was the fact that salt rejections fell rapidly when a critical level of
sulfonation was exceeded, but water fluxes were too low until this sulfonation
level was exceeded.
Beginning in late 1976, Cadotte and coworkers developed a composite sul-
fonated polysulfone membrane, using a slightly different approach.46 Polysul-
fone was sulfonated to a degree sufficient to produce complete water solubility.
Solutions of this polymer dissolved in aqueous or water-alcohol media were
coated onto microporous polysulfone sheets, then oven-dried at 100 to 14O’C.
This temperature treatment served to insolubilize the sulfonated polymer. In-
tramolecular sulfone crosslinks were assumed to be formed, though this was
never specifically proven. Additives such as polyols and polyphenols could be
added for additional crosslinking. These and other sulfonated membranes often
showed good rejection toward dilute salt solutions such as tapwater or 0.1%
aqueous sodium chloride. At increasing salinities, however, salt rejection levels
showed a rapid fall off. The composite sulfonated polysulfone membrane de-
scribed by Cadotte, for instance, exhibited 80% or less salt rejection in sea-
water reverse osmosis tests.
In sulfonated aromatic polymer membranes, therefore, Donnan ion exclu-
sion appears to be an important contributor to performance. In contact with
dilute feedwaters, these strongly anionic membranes exclude feedwater anions
from entering the membranes by charge repulsion effects, thus inhibiting salt
passage. In more concentrated salt solutions, the charge repulsion effects are
shielded by the high ionic activity of the feedwater.
A sulfonated polysulfone membrane with commercial potential has been
developed in the form of a hollow fiber by workers at Albany International
Corporation.86r87 In a 5,000-hour test on 3,500 ppm brackish water at 400 psi,
this membrane exhibited 98% salt rejection at 1 gfd. Flux and salt rejection re-
mained constant even with addition of 100 ppm chlorine. In a 12,000-hour test
on seawater at the Wrightsville Beach Test Facility, this membrane exhibited
98+% salt rejection and an average flux of 1.5 gfd at 1,000 psi operating pres-
sure. Thus, it is possible to make a high rejection membrane from sulfonated
polysulfone. Flux was rather low in this particular case, but suitable for hollow
fiber membrane use.
PLASMA POLYMERIZATION IN COMPOSITE MEMBRANE FABRICATION
Plasma polymerization has been explored in considerable detail as a poten-

tial route to composite reverse osmosis membranes. Yasuda has written an ex-
cellent review covering the fundamentals and progress of plasma polymerization
through 1974.s8 Advances since then have not been particularly promising from
a commercial standpoint, with the exception of the Solrox membrane (see be-
low).
Monomer polymerization by gas plasma techniques is quite complex, involv-
ing considerable fragmentation of vaporized monomers, and considerable cross-
linking and grafting of deposited films. Vinyl unsaturation is not necessarily re-
quired for monomers to be polymerizable. Plasma polymers generally have short
chain lengths, coupled with a high degree of branching and crosslinking, and con-
taining a high concentration of residual free radicals. Monomers containing oxy-
gen-based functional groups (such as carboxylic acids, ketones and esters) lose
much of their oxygen content during polymer deposition. Nitrogen-containing
monomers such as amines, on the other hand, show high levels of nitrogen in-
corporation into the deposited polymer films. Nitrogen gas and water vapor,
added to the gas plasma of an organic monomer, tend to show finite rates of
incorporation into the polymer deposits. To obtain good reverse osmosis char-
acteristics, sufficient polymer must be deposited to “plug” the holes on the sur-
face of a microporous substrate. Substrates having surface pore sizes of 200 to
300 angstroms appeared to be optimum in this regard.
Early work by Buck and Davar” examined the plasma polymerization of
several monomer systems. Best results in terms of reverse osmosis performance
were achieved with vinylene carbonate/acrylonitrile and vinyl acetate/acrylo-
nitrile.
Wydeven and his coworkers examined the plasma polymerization of several
amine monomers and obtained the best reverse osmosis membranes using allyl-
amine.g0r91‘Reverse osmosis performances of 98 to 99% salt rejection and 4 to
8 gfd flux were achieved at test conditions of 1.0% sodium chloride, 600 psi,
20°C. ESCA spectra showed the nitrogen groups in the plasma-formed polymer
to be nitrile or imine groups, but not amine groups. Oxygen was also present in
the ESCA analysis. Elemental analysis showed a membrane stoichiometry of
CsHs.sNo.&o.~.
Yasuda obtained membranes with good reverse osmosis properties by poly-
merizing gas plasma mixtures of acetylene/water/nitrogen and acetylene/wa-
ter/carbon monoxide.“r9j Elemental analysis of the acetylene/water/nitrogen
membrane was not given, but it was likely approximately equivalent to the allyl-
amine-derived composition. Yasuda has observed membrane performance levels
as high as 99% salt rejection and 38 gfd, tested on 3.5% sodium chloride at 1,500
psi.
To date, no commercial examples of these membrane compositions have
appeared in the marketplace.
SOLROX 0200 MEMBRANE
Sano and coworkers at Sumitomo Chemical Company developed the Solrox

membrane, which is technically a cross between asymmetric and composite

membrane technologies.B4* An asymmetric polymer membrane is formed
from a polyacrylonitrile copolymer. This membrane contains discrete micro-
pores, being more like an ultrafiltration membrane rather than a reverse osmosis
membrane. It is dried, then exposed to a helium or hydrogen gas plasma. The
plasma treatment forms a tight, crosslinked surface skin on the asymmetric poly-
acrylonitrile support. Some vaporization and redeposition of the film compo-
nents most likely occurs in this step. When placed on test, this plasma-modified
membrane develops good salt rejection and flux in about 24 hours of continu-
ous use. Analogs were also prepared using an organic monomer such as 4-vinyl-
pyridine in the gas plasma. Data in a 1979 patent show a water flux of 87 gfd
at 140 psi for the unmodified asymmetric film substrate.” After plasma treat-
ment, this film exhibited 11 gfd and 98.3% salt rejection on a 0.55% sodium
chloride feedwater at 700 psi.
Table 5.7, taken from a July 1982 technical bulletin by Sumitomo Chem-
ical, lists performance data for a series of Solrox membrane grades that can be
made by this process. Of these, the Solrox SC-0200 has been manufactured in
tubular form on a semicommercial basis and used within the Sumitomo organiza-
tion on organic chemical concentration applications. Table 5.8 lists organic re-
jections for the SC-0200 membrane.
These organic rejections, while greatly exceeding the capabilities of cellu-
lose acetate membranes, are not appreciably different than rejection levels ex-
hibited by aromatic polyamide membranes (FT-30, Permasep B-9). Nor do they
match the organic rejection characteristics of the PEC-1000 membrane. The Sol-
rox membrane is not resistant to chlorine, and its water flux is somewhat low
(about 25 gfd at 650 psi net driving pressure). Consequently, it has not become a
significant competitive membrane in the world marketplace.
Table 5.7: Available Types of Solrox Membranes
Reverse Osmosis Performance*

Salt Rejection Flux
Membrane Type (XI .(gfd)
SC-0100 99.0+ 18
SC-0200 97.5-98.5 27-32
SC-0500 94-96 35-41
SC-1000 88-92 41-47
SC-2000 75-85 50-71
* Test conditions: 0.5% NaCl, 710 psi, 25°C

Table 5.8: Organic Solute Rejections of SOLROX SC-0200 Membrane in

Comparison with Cellulose Acetate Membrane
Solrox SC-0200 Cellulose Acetate
Solute (X rej.)* (X rej.)*
Sodium chloride 58 98 98
Methanol 32 15 5
E than01 46 61 10
n-Propanol 60 79 25
n-Butanol 74 90 15
set-Butanol 74 97 35
Isobutanol 74 97 40
tert-Butanol 74 99 80
Ethylene glycol 62 78 __
Glycerol 92 98 __
Phenol 94 09 0
Ethyl cellosolve 90 96 -
Acetone 58 82 21
Urea 60 57 35
Acetic acid 60 51 25
Adipic acid 146 88 __
Aniline 93 94 8
*Test conditions: 1.0% aqueous solution, 710 psi, 25-C
MISCELLANEOUS COMPOSITE REVERSE OSMOSIS MEMBRANES
The emphasis in this chapter has been on commercially significant mem-

branes. Historical technical events leading to the successful development of these
membranes, and information on closely related membranes appearing in the
technical literature have also been presented. Other composite membranes have
appeared in the technical and patent literature, that have been omitted for lack
of time and space. Perusal of the U.S. government-supported research project
reports published by Cadotte, for instance, will reveal composite membranes
based on polyacrolein, styrene-maleic anhydride condensates, metaphenylene-
diamine-formaldehyde polymers, hydrophilic aromatic polyesters, sulfonated
polymers, terephthaloyl oxalic-bis-amidrazone polymers, and crosslinked poly-
vinyl alcohol derivatives. A few other interesting examples are given below, il-
lustrating the versatility of the composite membrane approach.
Strathmann prepared an all-polyimide composite membrane-both bottom
and top layers.” A microporous asymmetric film of the polyamic acid intermed-
iate was cast by quenching in acetone, then dried and thermally cyclized to the
polyimide at 300°C. The microporous polyimide sheet was then overcoated with
a dilute solution of the same polymer, which was allowed to evaporate to give a
300-angstrom-thick coating. This was also cyclized to the polyimide to generate
a fully solvent resistant reverse osmosis membrane.
Taketani and coworkers at Teijin have sought to produce a stable flux poly-
benzimidazolone (PBIL) membrane by producing it as an ultrathin coating on
microporous polysulfone.g* The key to this effort was the discovery that the
PBIL polymer was soluble in water containing 5% or more of a water-miscible
aliphaticamine. Thus, thesolvent sensitivity of polysulfone could be sidestepped.
Koyama, Okada and Nishimura prepared composite reverse osmosis mem-
branes containing an interpenetrating polymer network of polyvinyl alcohol
with poly(styrenesulfonic acid).” The barrier layer was insolubilized by heating
at 12O’C for 2 hours. Salt rejections in the low 90’s and rather low fluxes were
observed. The long time length of heat-curing at 12OOCwas probably a factor in
producing low fluxes. Kawahara and Yasuda obtained composite reverse os-
mosis membranes by acid-catalyzed crosslinking of methylolated poly(4-vinyl-
phenol) coatings on polysulfone substrates.rea
ADVANTAGES OF THE COMPOSITE MEMBRANE APPROACH
The numerous examples of composite membranes that have appeared in the

technical literature exemplify the versatility of this approach. To the chemist
attempting to develop new reverse osmosis membranes, the composite approach
has three major advantages.
First, large quantities of monomers or polymers are not required to fabri-
cate the barrier layer. At surface thicknesses of 200 to 2000 angstroms, roughly
0.02 to 0.2 grams of barrier layer polymer per square meter of membrane is re-
quired (excluding excess chemicals used in membrane fabrication). These are
orders of magnitude smaller than required in asymmetric membrane manufac-
ture. Thus, very expensive chemicals are not a deterrent to composite membrane
fabrication. Furthermore, one avoids the need for expertise in producing large
quantities of high polymers in a reproducible manner. Small companies that are
not basic in plastics production can emerge and compete effectively in this mem-
brane market.
In this context, only two polymers have ever been used on a large scale in
asymmetric membranes: cellulose acetate and Permasep B-g/B-IO aramids. The
former polymer predates the era of reverse osmosis membranes. The latter poly-
mer has been used in hollow fiber membranes for 15 years. Attempts to bring
other new polymers into asymmetric membrane production have been few
(PBIL, PBI, polypiperazineamides), generally without particular success.
Second, insoluble crosslinked barrier layer compositions are possible, and,
in fact, are almost universal in the composite membrane approach. Optjmum
reverse osmosis performance and chemical stability can be achieved, in part, due
to preparation of crosslinked compositions. This is readily possible by the com-
posite membrane approach, but not so simple by the asymmetric membrane ap-
preach. The PA-300, FT-30, and PEC-1000 barrier layer compositions, for ex-
ample, are simply not feasible to prepare by asymmetric film casting techniques.
The composite approach, therefore, is far more versatile.
Water flux through reverse osmosis membranes is considerably dependent
on the hydrophilic character of the barrier layer. In the composite membrane
approach, highly hydrophilic barrier layer compositions can be used, suitably
insolubilized by crosslinking. To a large extent, water flux and salt rejection
can be controlled by the type and extent of crosslinking.
Third, each layer of a composite membrane can be optimized. The barrier
layer can be developed to display single pass seawater desalination character-

istics, on the one hand, or to selectively pass monovalent salts, on theother hand.
The microporous substructure can be optimized for mechanical strength. The
combination can be optimized for maximum water permeation rates.
These advantages are reflected in the intense activity in filing patents on
composite reverse osmosis membranes, and in the growing selection of commer-
cially available types. Because of these advantages, many new examples of such
membranes are likely to reach commercial status in the future, whereas very few
new asymmetric membranes can be anticipated.
REFERENCES
1. Cadotte, J.E. and Petersen, R.J., “Thin-Film-Composite Reverse Osmosis

Membranes: Origin, Development, and Recent Advances”, in Synthetic
Membranes: Vol I. Desalination, A.F. Turbak, Ed., American Chemical
Society, Washington, D.C., 1981, p. 305.
2. Cadotte, J.E., “Evolution of Composite Reverse Osmosis Membranes”, in
Materials Science of Synthetic Membranes, D.R. Lloyd, Ed., American
Chemical Society, Washington, D.C., 1985, p. 273.
3. Francis, P.S., “Fabrication and Evaluation of New Ultrathin Reverse Os-
mosis Membranes”, NTIS Report No. PB-177083, available from Na-
tional Technical Information Service, U.S. Department of Commerce,
Springfield, VA 22161 (Feb. 1966).
4. Rozelle, L.T., Cadotte, J.E. and McClure, D.J., “Ultrathin Cellulose Ace-
tate Membranes for Water Desalination” in Membranes from Cellulose
and Cellulose Derivatives, A.F. Turbak, Ed., Interscience Publishers, New
York, 1970, p. 61.
5. Rozelle, L.T., Cadotte, J.E., Senechal, A.J., King, W.L. and Nelson, B.R.,
“Tubular Ultrathin Membranes for Water Desalination” in Reverse Os-
mosis Membrane Research, H.K. Lonsdale and H.E. Podall, Ed., Plenum
Press, New York, 1972, p. 419.
6. Rozelle, L.T., Cadotte, J.E. and McClure, D.J., “Development of New Re-
verse Osmosis Membranes for Desalination”, NTIS Report No. PB-
201016, lot. cit. (June 1970).
7. Rozelle, L.T. and Petersen, R.J., “Development of a New Concept in Mem-
brane Structure for Application in Hemodialysis”, Sixth Ann. Prog.
Report, NTIS Report No. PB-230997, lot. cit. (Jan. 1974).
8. Rozelle, L T. and Petersen, R.J., “Ultrathin Membranes for Blood Oxy-
genators”, Second Ann. Prog. Report, NTIS Report No. PB-231324,
lot. cit. (Jan. 1974).
9. Forester, R.H. and Francis, P.S., U.S. Patent 3,551,244 (Dec. 29, 1970).
IO. Ward, W.J. III, U.S. Patent4,374,891 (Feb. 22, 1983).
11. Salemme, R.M. and Browall, W.R., U.S. Patent 4,155,793 (May 22, 1979).
12. Tolansky, S., An Introduction to lnterferometry, Longmans Green, Lon-
don, 1965.
13. Nomarski, M.G.,J. Phys. Rad., 16, 95 (1955).
14. Carnell, P.H. and Cassidy, H.G.,J. Appl. Polym. Sci., 9, 1863 (1965).
15. Riley, R.L., Lonsdale, H.K., Lyons, C.R. and Merten, U., J. Appl. Polym.
Sci., 11, 2143 (1967).
16. Petersen, R.J. and Johnston, P., “Evaluation of Nonthrombogenic Fluoro-
polymer Membranes in Blood Oxygenators”, NTIS Report No. PB-
251585, lot. cit. (Feb. 1976).
17. Riley, R.L., Lonsdale, H.K., La Grange, L.D. and Lyons, CR., “Develop-
ment of Ultrathin Membranes”, NTIS Report No. PB-207036, lot. cit.
(Jan. 1969).
18. Riley, R.L., Lonsdale, H.K. and Lyons, C.R., J. Appl. Polym. Scb, 15, 1267
(1971).
19. Cadotte, J.E. and Francis, P.S., “Development of New Reverse Osmosis
Membranes for Desalination”, First Quarterly Report on Contract No.
14-01-0001-1143 to Office of Saline Water, U.S. Department of the In-
terior; Desalination Abstracts, Vol l/2, Abstract No. C-395 (p. 133)
196611967.
20. Rozelle, L.T., Cadotte, J.E., Corneliussen, R.D. and Erickson, E.E., “‘Final
Report on Development of New Reverse Osmosis Membranes”, NTIS Re-
port No. PB-206329 (June 1968).
21. Michaels, A.S., U.S. Patent 3,615,024 (Oct. 26, 1971).
22. Cadotte, J.E. and Rozelle, L.T., “In-Situ-Formed Condensation Polymers
for Reverse Osmosis Membranes”, NTIS Report No. PB-229337, lot. cit.
(Nov. 1972).
23. Rozelle, L.T., Cadotte, J.E., Cobian, K.E. and Kopp, C.V., Jr., “Nonpoly-
saccharide Membranes for Reverse Osmosis: NS-100 Membranes”, in Re-
verse Osmosis and Synthetic Membranes, S. Sourirajan, Ed., National Re-
search Council of Canada, Ottawa, 1977, p. 249.
24. Cadotte, J.E., U.S. Patent 4,039,440 (Aug. 2, 1977).
25. Sudak, R.G., ‘Bott, J.B., Fox, R.L., Gagner, S.J. and McKee, M.E., “Devel-
opment of Reverse Osmosis Membrane for Seawater Desalination”, NTIS
Report No. PB80-182090, lot. cit. (Sept. 1979).
26. Milstead, C.E. and Riley, R.L., “Low Pressure Brackish Water Reverse Os-
mosis Membrane Development”, NTIS Report No. PB80-153539, lot.
cit., (Feb. 1980).
27. Wrasidlo, W.J., U.S. Patent 3,951,815 (April 10, 1976).
28. Fang, H.H.P. and Chian, E.S.K.,J. Appl. Polym. Sci., 20, 303 (1976).
29. Fang, H.H.P. and Chian, E.S.K., Environ. Sci. Technol., 10, 364 (1976).
30. Chian, E.S.K., Bruce, W.N. and Fang, H.H.P., Environ. Sci. Technol., 9, 52
(1975).
31. Wrasidlo, W.J., U.S. Patent 4,005,012 (Jan. 25, 1977).
32. Riley, R.L., Fox, R.L., Lyons, C.R., Milstead, C.E.,Seroy,M.W.andTagami,
M., Desalination, 19, 113 (1976).
33. Riley, R.L., Milstead, C.E., Lloyd, A.L., Seroy, M.W. and Tagami, M., De-
salination, 23,331 (1977).
34. Leban, M.I. and Wydeven, T.J., Environ. Sci. Technol., 18, 778 (1984).
35. Hickman, C.E., Jamjoom, I ., Riedinger, A.B. and Seaton, R.E ., Desalination,
30, 259 ( 1979).
36. Kutbi, I.I., Metwally, A.M., Zabri, Z.A. and Husseiny, A.A., Desalination,
39, 179 (1981).
37. Riley, R.L., Case, P.A., Lloyd, A.L., Milstead, C.E. and Tagami, M., De-
salination, 36, 207 (1981).
38. Kurihara, M., Uemura, T. and Okada, K., U.S. Patent 4,387,024 (June 7,
1983).
39. Kawaguchi, T., Taketani, Y., Minematsu, H., Sasaki, N., Hayashi, Y. and
Hara, S., U.S. Patent 4,242,208 (Dec. 30, 1980).
40. Kawaguchi, T., Taketani, Y., Hayashi, Y., Ono, T. and Mori, K., U.S. Patent
4,265,745 (May 5, 1981).
41. Kawaguchi, T., Taketani, Y., Sasaki, N., Minematsu, H., Hayashi, Y. and
Hara, S., US. Patent 4,302,336 (Nov. 24, 1981).
42. Kawaguchi, T., Minematsu, H., Hayashi, Y. and Hara, S., U.S. Patent
4,388,189 (June 14, 1983).
43. Kawaguchi, T., Minematsu, H., Hayashi, Y., Hara, S. and Ueda, F., U.S.
Patent 4,360,434 (Nov. 23, 1982).
44. Yaginuma, H., U.S. Patent 4,244,817 (Jan. 13, 1981).
45. Cadotte, J.E., Cobian, K.E., Forester, R.H. and Petersen, R.J., “Continued
Evaluation of In Situ-Formed Condensation Polymers for Reverse Os-
mosis Membranes”, NTIS Report No. PB-253193, lot. cit. (Apr. 1976).
46. Cadotte, J.E., Steuck, M.J. and Petersen, R.J., “Research on In Situ-Formed
Condensation Polymers for Reverse Osmosis Membranes”, NTIS Report
No. PB-288387, lot. cit. (March 1978).
47. Cadotte, J.E., King, R.S., Majerle, R.J. and Petersen, R.J., .I. Macromol.
Sci. Chem., A15, 727 (1981).
48. Cadotte, J.E., U.S. Patent 4,259,183 (March 31, 1981).
49. Credali, L., Chiolle, A. and Parini, P., Desalination, 14, 137 (1974).
50. Parrini, P., Desalination, 48, 67 (1983).
51. Petersen, R.J., Cadotte, J.E. and Buettner, J.M., “Development of FT-30
Membranes in Spiral Wound Modules”, NTIS Report No. PB83-191775
lot. cit. (Oct. 1982).
52. Brady, M.F., Iwahori, H. and Murata, K., “New Thin-Film Composite Re-
verse Osmosis Membranes and Spiral Wound Modules”, Technical Pro-
ceedings of the 12th Ann. Conf. of the Water Supply Improvement
Assoc., Volume II: Capsule Presentations, Orlando, Florida, May 13-18,
1984, available from WSIA, P-0. 387, Topsfield, MA.
53. Kamiyama, Y., Yoshioko, N. and Nakagome, K., Ger. Offen. DE 3,129,702
(Feb. 11, 1982).
54. Ohya, H., Maku (Membrane), 10, 101 (1985).
55. Cadotte, J.E., U.S. Patent 4,277,344 (July 7, 1981).
56. Cadotte, J.E., Petersen, R.J., Larson, R.E., and Erickson, E.E., Desalination,
32, 25 (1981).
57. Koo, J-y., Reitz, J.M., Cadotte, J.E. and Petersen, R.J., “Nonfouling Coat-
ings for Thin Film Composite Membranes”, final report on Contract No.
14-08-0001-2288 submitted to U.S. Department of the Interior, in press.
58. Glater, J., McCutchan, J.W., McCray, S.8. and Zachariah, M.R., “The Ef-
fect of Halogens on the Performance and Durability of Reverse-Osmosis
Membranes”, .in Synthetic Membranes: Vol / Desalination, A.F. Turbak,
Ed., American Chemical Society, Washington, D.C., 1981, p. 171.
59. Bhattacharyya, D., Jevtitch, M., Ghosal, J.K. and Kozminski, J., Environ.
Prog., 3(2), 95 (1984).
60. Larson, R.E., Cadotte, J.E. and Petersen, R.J., “Development of the FT-30
Thin Film Composite Membrane for Seawater Desalting Applications,”
NWSIA Journal, Vol. 8, No. 1, p. 15 (Jan. 1981).
61. Larson, R.E., Cadotte, J.E. and Petersen, R.J., Desalination, 38, 473
(1981).
62. Larson, R.E., Petersen, R.J. and Eriksson, P.K., Desalination, 46, 81
(1983).
63. Federal Register, Vol. 49, No. 246, p. 49448 (Dec. 20, 1984).
64. Hedges, R.M. and Pepper, D., Filtration 8t Separation, March/April 1984,
p. 112.
65. Nielsen, W.K. and Kristensen, S., Proceedings-lnt. Conf. on Desalination
and Water Reuse, Vol. 1, p. 383 (1981); Desalination, 38 (11, 605
(1981).
66. Rak, S.F. and Ward, K., U.K. Patent Application 2,139,113A (Nov. 7,
1984).
67. Morgan, P.W., Condensation Polymers: By Interfacial and Solution Meth-
ods, Interscience Publishers, 1965, p. 22.
68. Fukuchi, S., Hayashi, T., Kobayashi, H. and Oshiumi, R., U.S. Patent
4,337,154 (June 29, 1982).
69. Cadotte, J.E., U.S. Patent 3,926,798 (Dec. 16, 1975).
70. Cadotte, J.E., Kopp, C.V., Jr., Cobian, K.E. and Rozelle, L.T., “In Situ-
Formed Condensation Polymers for Reverse Osmosis: Second Phase”,
NTIS Report No. PB-234198, lot. cit. (June 1974).
71. Cadotte, J.E., Cobian, K.E., Forester, R.H. and Petersen, R.J., “In Situ-
Formed Condensation Polymers for Reverse Osmosis Membranes”,
NTIS Report No. PB-248670, lot. cit. (Feb. 1975).
72. Graefe, A.F., “Development of a Composite Reverse Osmosis Membrane
for Single Pass Seawater Desalination”, NTIS Report No. PB80-124852,
lot. cit. (Feb. 1979).
73. Graefe, A.F. and Schell, W.J., “Development of Low Pressure Brackish
Water Reverse Osmosis Membranes”, NTIS Report No. PB80-126204,
lot. cit. (May 1979).
74. Schiffer, D.K., Davis, R.B. and Coplan, M.J., “Development of Composite
Hollow Fiber Reverse Osmosis Systems”, NTIS Report No. PB80-213044,
lot. cit. (June 1979).
75. Kurihara, M., Kanamaru, N., Harumiya, N., Yoshimura, K. and Hagiwara,
S., Desalination, 32, 13 (1980).
76. Kurihara, M., Watanabe, T. and Inoue, T., U.S. Patent 4.366.062 (Dec. 28,
1982).
77. Kurihara, M., Harumiya, N., Kanamaru, N., Tonomura, T. and Nakasatomi,
M., Desalination, 38,449 (1981).
78. Kurihara, M., Nakagawa, Y., Takeuchi, H., Kanamaru, N. and Tonomura,
T., Desalination, 46, 101 (1983).
79. Kurihara, M., Maku (Membrane), 8 (2). 97 (1983).
80. Chen, J-y., Kurihara, M. and Pusch, W., Desalination, 46, 379 (I 983).
81. Kamiyama, Y., Yoshioka, N., Matsui, K. and Nakagome, K., Desalination,
51,79 (1984).
82. Plummer, C.W., Kimura, G. and La Conti, A.B., “Development of Sul-

fonated Polyphenylene Oxide Membranes for Reverse Osmosis”, NTIS
Report No. PB-201034, lot. cit. (May 1970).
83. La Conti, A.B., “Advances in Development of Sulfonated PPO and Modi-
fied PPO Membrane Systems for Some Unique Reverse Osmosis Applica-
tions”, in Reverse Osmosis and Synthetic Membranes, S. Sourirajan, Ed.,
lot. cit., p. 21 1.
84. Quentin, J-P., U.S. Patent 3,709,841 (Jan. 9, 1973).
85. Bourganel, J., U.S. Patent 4,026,977 (May 31, 1977).
86. Schiffer, D.K. and Coplan, M.J., “Development of Hollow Fiber Reverse
Osmosis Membranes”, NTIS Report No. PB81-167215, lot. cit. (Nov.
1980).
87. Davis, R.B., Schiffer, D.K. and Kramer, C.E., “Hollow Fiber Reverse Os-
mosis Composite Membranes: Process and Properties”, in Synthetic
Membranes: Vol 1. Desalination, A.F. Turbak, Ed., American Chemical
Society, Washington, D.C., 1981, p. 367.
88. Yasuda, H., “Composite Reverse Osmosis Membranes Prepared by Plasma
Polymerization”, in Reverse Osmosis and Synthetic Membranes, S.
Sourirajan, Ed., lot. cit., p. 263.
89. Buck, K.R. and Davar, V.R., Br. Polymer J., 2,238 (1970).
90. Bell, A.T., Wydeven, T. and Johnson, C., J, Appl. Polym. Sci., 19, 1911
(1975).
91. Hollahan, J.R. and Wydeven, T., J. Appl. Polym. Sci., 21, 923 (1977).
92. Yasuda, H., Marsh, H.C. and Tsai, J., J. Appl. Polym. Sci., 19, 2157
(1975).
93. Yasuda, H. and Marsh, H.C.,J. Appl. Polym. Sci., 19.2981 (1975).
94. Sano, T., Shimomura, T., Sasaki, M. and Murase, I., U.S. Patent 4,107,049
(Aug. 15, 1978).
(April 3, 1979).
(June 9, 1981).
97. Strathmann, H., U.S. Patent 4,071,590 (Jan. 31, 1978).
98. Taketani, Y., Kawaguchi, T., Ono, T. and Mori, K., U.S. Patent 4,260,652
(April 7, 1981).
99. Koyama, K., Okada, M. and Nishimura, M., J. Appl. Polym. Sci., 27,
2783 (1982).
100. Kawahara, H. and Yasuda, T., U.S. Patent 4,326,958 (April 27, 1982).
6
Process Design and Optimization
Robert Rautenbach
INTRODUCTION-MASS TRANSPORT AT THE MEMBRANE SURFACE
The Local Mass Transport

It is understood that the economical success of any membrane process de-
pends primarily on the quality of the membrane, specifically on flux, selectivity
and service lifetime. Consideration of only the transport mechanisms in mem-
branes, however, will in general, lead to an overestimation of the specific per-
meation rates in membrane processes. Formation of a concentration boundary
layer in front of the membrane surface or within the porous support structure
reduces the permeation rate and, in most cases, the product quality as well.
For reverse osmosis, Figure 6.1 shows how a concentration boundary layer
(concentration polarization) forms as a result of membrane selectivity. At steady
state conditions, the retained components must be transported back into the
bulk of the liquid. As laminar flow is present near the membrane surface, this
backflow is of diffusive nature, i.e., is based on a concentration gradient. At
steady state conditions, the concentration profile is calculated from a mass bal-
ance as
‘I,/ k
(1) ‘2-‘3 = e
9
c1-c3
where k is the mass transfer coefficient which can be assumed, at least with a
good approximation, to be independent of the permeate flux V,. For this rea-
son, the analogy between heat and mass transfer is valid and k can be calculated
using the well-known heat transfer equations. This has been tested experimen-
tally for a number of cases.’
349
a
n-r3 mass transport by
convection
permeate
diffusive mass transport
c3 F membrane
Figure 6.1: Concentration polarization.
Figures 6.2 and 6.3 indicate the order of magnitude of concentration polar-
ization for laminar and turbulent flows through tubular membranes. The dia-
grams illustrate the dependence of the concentration boundary layer on flow
conditions along the membrane (Re) and on the permeation flux (Pe,).
II
=-
v,.d
D i
100 1000
Reynolds number Re
Figure 6.2: Concentration polarization in laminar flow.
Process Design and Optimization 351
1
“I
OS VI7I.d ! !!
- Pex-fj-
- Sh= ~.02JRe7/8!k”L
400 .\
Reynolds number Re
Figure 6.3: Concentration polarization in turbulent flow.
According to Figures 6.2 and 6.3, turbulent flow is more advantageous. Fur-
thermore, these diagrams explain why, in the case of gas permeation, in contrast
to separation processes in the liquid state, the resistance due to a boundary layer
can be neglected: the diffusion coefficients of gases, for example 02/Nz are lo4
times larger than those of dissolved components in liquids such as NaCI/l$O
which, in combination with the permeation rates reported in literature, leads
to Pe numbers of less than unity.
Influence of the Asymmetric Structure of Membranes

While it is possible to enhance mass transfer in the boundary layer by im-
proving the flow conditions, this is impossible if a concentration profile exists
within the porous support structure.
For reverse osmosis, Figure 6.4 shows the predicted concentration profile
in the porous structure of an asymmetric membrane. According to Figure 6.4,
in liquid systems, the concentration c3, which affects permeate flux and product
quality, cannot be influenced significantly by the concentration of the liquid on
the product side. Conditions are nearly always favorable for the permeate to flow
unhindered. Even in the extreme case of local concentration on the permeate
side being equal to that of the liquid on the high pressure side of the membrane
(c, = c,), the error made in calculating the solvent flux V, and the salt flux J,
is only 2% or 5% respectively, if the effect of cs is neglected.
Again, the situation is different in the case of gas permeation. Here, the lo-
cal concentration affects the (local) permeate fluxes of the mixture components
and, for this reason, the type of flow for gas permeation on the permeate side is
much more important than for liquid separation processes such as reverse os-
mosis. Figure 6.5 shows theoretical results of Blaisdell and Kammermeyer.2
Mol fraction of oxygen is plotted against the yield during fractionation of air
for different flow configurations and cases of mixing. The diagram demonstrates
that (1) countercurrent flow provides the best results and (2) marked differences
exist between the perfect mixing and plug flow.
feed ~membrone-----_tcpermeate
porous
subslucture
membrane thickness
Figure 6.4: Concentration profile in porous substructure of asymmetric mem-

branes.
0.29, t 1 1 I
I
0.2A - complete mixing
stage-cut 0
Figure 6.5: Effect of flow configuration in gas permeation.

Change of Conditions Along the Membrane

As a consequence of the permeate flux, the pressure gradient as well as the
mean velocity and concentration will vary along the membrane surface-in prin-
ciple at the feed side as well as at the product side. Depending on module de-
sign, this has to be taken into account for one or both sides. If, for example, the
membrane is of tubular design with feed-flow inside, the following has to be ex-
pected along the tube-axis:
(1) Decrease of the transmembrane pressure-difference as a conse-

quence of the drop in pressure due to friction ( + decrease of per-
meate flux).
(2) Increase of the mean concentration of the feed solution ( + in-
crease of osmotic pressure + decrease of permeate flux and in-
crease of salt flux).
(3) Decrease of the mean velocity ( + increase of concentration polar-
ization + decrease of permeate flux and increase of salt flux).
For these reasons, the overall permeation rate and the product quality of a
membrane channel have to be calculated by integrating over the length elements
employing, in addition to the equations discussed above,
(1) the mass balance

(2) the material balance
(3) the energy balance.
The equations necessary for the calculation of tubular RO-membrane chan-

nels (flow of permeate unhindered, no influence of the porous support-layer) are:
mass transfer at the membrane surface
concentration polarisation e I evw’,,

I 3
equation for mass transfer Sh = y = ShlRe, Sc)
mass transfer in the membrane

permeate flux V, = A,IAp-blcz-c,)) ; AP ‘P, -P,,,
salt flux 3s = WC,-c,)
salt concentration C) = Ir
VW
balance for a length element
LAX
mass balance v,*,., = K..- - (j VW*,
material balance (salt) ‘jr.“., CI,~.( =TX,n.Cl,n- 9 .V,,. C,,,
x,d LAX.‘&
energy balance(p=constl pm+,= &[p”+~pT,,~(1-5 l-~.Pn
- + P %:,,.I - APV
This set of equations has to be solved numerically.

Figure 6.6 discusses the results of such a numerical calculation of a long tu-
bular RO-membrane channel for seawater desalination. It clearly demonstrates
the decrease of permeate (mass) flux J, and, at the same time, the increase of
local and mean product concentration cs and &.
0
‘0 20 LO 60 80 100
Number of tubes
Figure 6.6: Calculation of a tubular module.
MODULE CONCEPTS AND DESIGN
The module, i.e., the industrial configuration of membranes, has to meet a

variety of requirements which are sometimes even contradictory. Points of major
importance are: (1) flow conditions along the membranes, (2) ratio of membrane
area to pressurized vessel volume, (3) price of module and (4) the possibility of
cleaning the membrane.
Depending on the process application, one or the other of these require-
ments is of primary importance and, for this reason, a number of different mod-
ules have been designed. The most important designs are the plate and frame,
the spiral wound and the hollow fiber module. Optimization procedure and
some of its results will be discussed for only one module configuration, the hol-
low fiber module-and for only one application-RD.
The Hollow Fiber Module

Hollow fiber modules contain very fine fibers forming asymmetric or sym-
metric membranes and capable of withstanding pressure differences (high-pres-
50- 10
bar l$ $!
-6 0 20 LO 60 60 100 cm 120
fibre length L
Figure 6.6: Permeation behaviour of hollow fibers.
fibre length L
Figure 6.9: Concentration profiles in hollow fibers.

In theory, larger fiber diameters should increase the flux, i.e., the perform-
ance of a single fiber. However, at the same time, the membrane area per unit
volume decreases. For some simplifying assumptions, it can be easily discussed
how fiber diameter and fiber length have to be chosen in order to maximize the
“yield” per unit volume of the bundle.
Neglecting (1) the radial pressure losses in the feed and (2) the influence of
osmotic pressure on flux, the specific yield of the fiber bundle is given by:
$2$&d&.
v
4(1-E)
d
3 3
(8) tanh (HLI

l+Hsls*tanh (HL)
For a fixed ratio of 2 = 2, Equation 8 is discussed in Figure 6.10.

I
I I I I I 1
0 20 LO 60 80 plm 100
fibre diameter di
Figure 6.10: Optimum design of hollow fibers.
According to Figure 6.10, an optimal fiber length exists for a given fiber
diameter. Figure 6.8 clearly indicates that hollow fiber modules for reverse os-
mosis should be made as short as possible. It should be noticed that a simple
correlation between yield of the single fiber and yield of the bundle can be de-
rived in any case only if the concentration and the pressure in the shell (outside
of the bundle) are considered to be constant! In principle, the concentration of
the feed increases between entrance and outlet while the pressure of the feed
decreases. A numerical solution taking this into account can be achieved if the
bundle is considered as a continuum containing sinks (Figure 6.11). In this case,
the following equations have to be solved:
mass balance (pL = const):
(9) = -& (r V, (t-11 =
material balance:
energy balance:
dp, (r) U-EIZ ; (r)

(11)
-a-i=-= - K0 nFtv] Ej 1
K, has to be varied according to the fiber arrangement. K, = 30

is valid for fibers in parallel.
concentrate
fiber bundle
on
permeate 1
Q39C30 1
feed
solution
Qla, ClorlPISX
Figure 6.11: Balance element in the fiber bundle.
In the case of reverse osmosis, the relative flow configuration does not af-
fect the performance to any large extent. As already indicated, the situation is
quite different for gas permeation. While countercurrent flow improves the sepa-
ration efficiency of hollow fiber modules in reverse osmosis only slightly, as far
as gas permeation is concerned, countercurrent flow produces the best results. A

consequent application of the countercurrent principle led to the concept of
Hwang,4 i.e., the membrane column (Figure 6.12). This column consists of a rec-
tification and a stripping section and permits binary mixture to be fractionated
into its components. For a given geometry of the membrane fractionating col-
umn (length of each section, membrane area per unit length), the product qual-
ity is determined by the flow rate of the gas being recirculated.
product 1
(permeate I
4
i
I enriching section
I
I
I
feed
I
I stripping section
I
I
I
product 2
(concentrate 1
Figure 6.12: Membrane column for gas separation.
The modules employed in a membrane fractionating column must be dif-

ferent from those used in cascades: in hollow fiber modules, for example, the
fibers cannot be U-shaped and must be open at both ends. In Reference 5, a
module is described based on this principle.
CASCADES
There are cases where it is impossible to fractionate a given mixture in one

stage to the desired degree of purity; the separation process will consist of sev-
eral stages. These stages can be realized in several ways, for example, by employ-
ing the countercurrent flow principle as much as possible within one unit like in
packed towers for absorption or rectification or by combining definite separa-
tion units in the form of a cascade. The best known examples of this case are
fractionating towers employing trays.
In fractionating towers, the trays are still incorporated in one vessel, i.e.,
part of one unit. In other cases, for example, cascades of hydrocyclones, gas
centrifuges and-last not least-membrane modules, the unit itself will be the stage
of the cascade.
In membrane technology, processes consisting of more than one stage are
known in gas permeation and for the production of boiler feed water from sea-
water. Furthermore, cascades will be necessary if organic mixtures are to be sep-
arated by membrane processes only.
The best known example of a membrane cascade consisting of a high num-
ber of stages is the Oak-Ridge plant for enriching U-235 in the gaseous phase
(employing porous membranes).
Definitions
Separation Unit. The elements of a separation process. Separation units of
a separation process could be, for example, a gas centrifuge, a membrane mod-
ule, a tray of distillation column or the evaporator of a multiple effect plant. It
should be kept in mind, however, that in plate-and-frame modules as well as in
modules of the spiral-wound-type usually every block or pressure-vessel contains
more than one separation unit!
Stage. The element of a cascade will be named “stage” and the stage itself
might contain several separation units connected in series and/or in parallel. Sep-
aration units are considered part of a stage as long as they are connected only at
the feed and/or retentate side!
The elements of a stage can be arranged according to Figure 6.13 (this is
sensible with respect to mass transfer in the boundary layer because of the de-
creasing volume flow at the feed side along the “axis” of a stage), in a “tapered”
fashion or in a “squared-off” fashion.
Figure 6.13: “Tapered” module arrangement.

P(Product1 , yp
Rnlxn
.
,,bed& RN ‘,“N
(Retentatel
Figure 6.14: Cascade without reflux.
Cascade. A combination of stages where the permeate of a stage will be

the feed of the next stage.
Yield. Yield is defined as the ratio of a certain (key) component in the
product and in the feed:
(12)
Selectivity. Selectivity in general is defined as the quotient of ratios of con-

veniently chosen concentration measures. Convenient measures for a concentra-
tion are, for example, the partial pressure pi and the molar fraction xi or a mass
fraction Wi. Accordingly, the selectivity is defined as:
PA’PB 1 Prod = ‘A”BI Prod

(13) s, 5 /-
PA PBI Feed ‘A’“B 1 Feed
or
WA”JB 1 Prod
(14) s2 = -i--7?-
A B1 Feed
For the permeation of ideal gases, where

.
n.
1 = Qi (PI xi - P2 Yi )
describes the transport mechanism (in sorption-diffusion) membranes, Equation

13 is reduced to
(15)
s=-=;
QA
,O<ol<l
QB
(in cases, where psvi << pr Xi and, consequently, ni - Qiprxi is valid).

Another entity which is often employed because of its simplicity is defined
as:
(16). x =-'AI Permeate = yA

x
xAj Retentate A
or
xn =
YAPI module p ermeate, stage n =-Yn
'AR1 module retentate, stage n xn
if characterizing the selectivity of the entire module.

“Cut-Rate” (Split-Factor). The cut-rate, sometimes called “split-factor”,
is defined as the ratio of permeate flow and feed flow. There will be, however,
a difference between the cut-rate defined for molar flows and for mass flow.
(17)
or
(18)
Operating Line and “Equilibrium” Curve. Both terms are of importance for
the graphical solution of a separation problem, i.e., for the graphical determina-
tion of the number of stages of a cascade. This method has been developed for
the design of distillation columns by MacCabe and Thiele and should be well
known. For all cases, the operating line represents the mass and material bal-
ances. In distillation, the equilibrium curve represents the thermodynamical va-
por/liquid equilibrium. For an ideal binary system, the equilibrium curve can be
calculated from Raoult’s law and the saturation-pressure curves of the pure com-
ponents of the mixture. In all other cases, however, for example, for all mem-
brane processes, the equilibrium curve does not represent a thermodynamical
equilibrium at all but will represent the separation characteristics of the module
or that of the stage.
CascadesWithout Reflux
Such cascades are only sensible in cases where the retentate is practically
worthless. Norsk-Hydro, for example, is operating an electrolysis-heavy-water-
production-plant where the deuterium is separated in a cascade without reflux
(at least in the lower part of the cascade). Figure 6.14 illustrates the principle
of a cascade without reflux.
Reflux Cascades
The inevitable losses of product with the retentate of cascades without re-
flux can be avoided by recycling the retentate according to Figure 6.15. This ad-
vantage, however, has to be weighed against the higher capital costs of reflux
cascades. Only a detailed calculation of the specific separating costs will lead to
a sound decision on whether a separation process should operate with or without
reflux. In principle, there are four possible modes of operation:
(I) Constant reflux R/P

(2) Constant cut rate and variable reflux
(3) Variable cut rate and variable reflux
(4) “Ideal” cascade operation (x, _ t,R = x, + 1,p).
Here, only the first two cases will be discussed because with modern computers
at hand, for these two cases only a graphical solution will be of value.
Constant Reflux. In general, the operating line of a cascade follows from a
mass/molar balance and a material balance between the top and a certain stage n
of the cascade. For constant reflux R = const resp. R/P, = v = const.
the molar balance:
P - Rn - p, = 0
n+l
and the material balance:
(20) P
yn+l -
Rnxn P, y, = g
n+l
result in:
V 1
(21) Yn+l = v+l % G-i y1
The operating line is a straight line with the slope --$
If the “equilibrium curve” is known, the number of stages necessary to

achieve a certain product quality can be easily found according to Figure 6.16
by a step-by-step graphical procedure.
Constant Cut Rate, Variable Reflux. When the molar balance:
(22) P - Rn - P, = 0
n+l
is developed for several stages, a pattern can be noticed if the (constant) cut
rate is introduced:
Rn,xn
RN-l
Pm,Ym
Rm-loxm-l_, -5--
--
RM-l ‘“M-l
Figure 6.15: Refl ux cascade.
Figure 6.16: Graphical evaluation of the number of stages for a reflux cascade
(enriching section), operated with constant reflux.
(23) = P, + R, = P, (l+y)
p2
Here, y = R/P is used instead of 0 = P/F. 8 and y are related by the molar bal-
ance of a stage: y = (1 - 0)/e.
= P, + R2 = P+l+Y (l+y))
p3
pn = P, Cl+ Y +y2 + . . . Y”“)
(24) 1 -y”
‘n =pl -1 -y
With the material balance:
Rn pl Yl
(25) y n+l =Gxn +p,,1
the operating line follows:

n+l
(26)
_ Y -Y X 1-Y
yn+ 1 n+l n +n+r Yl
1-Y 1 -Y
Quite obviously, the slope of the operating line differs from stage to stage! The
number of stages follows according to Figure 6.17. It should be noticed that all
operating lineswill intersect at y1 = x1.
Figure 6.17: Graphical evaluation of the number of stages for a reflux cascade
(enriching section), operated with constant cut-rate, variable reflux.
Once the necessary number of stages is known, the product rate PI/F can be
calculated according to:
(27)
The “Equilibrium Curve”

As mentioned above, the “equilibrium curve” in the case of membrane cas-
cades is not at all a thermodynamic equilibrium but will represent the separation
characteristics of the module or of the stage. This separation characteristic is in
general a function of:
(1) the selectivity of the membrane

(2) the fluid dynamics in the module
(3) the driving force
(4) the concentration level (because the membrane selectivity is or
might be concentration dependent)
(5) the flow pattern in the module, for example, co- or countercurrent
flow.
With respect to flow pattern, five different patterns are possible:
(1) ~~z;r?$e mixing at both sides of the membrane (Weller Steiner
(2) Cross-plug flow, i.e., plug flow at the feed side and permeate flow
orthogonal to the membrane without mixing (Weller Steiner Case
2 6 Naylor-Baker’)’
(3) Parallel plug flow, cocurrent flow (Blaisdell-Case’)
(4) Parallel plug flow, countercurrent flow
(5) Partial (incomplete) mixing of feed and permeate (Breuer Case’ ).
Here, only the first case will be discussed. A detailed discussion of the other
cases can be found elsewhere.lO~‘l
Complete Mixing of Feed and Permeate. In this case (Figure 6.18), the ma-
terial balances for a binary mixture are:
P* Y * = Q, (P, x, - ?2 y*)
(29) P,(l- Y n) = Q, (p*(l-x, > - p2(1- Yn))
Combining of both equations and introducing 6 = ps/pr and o = QB/QA leads to:
n
(30)
;‘-I- = L
-‘pn o ( Ynx
-.
n
-1 -(l-
yn c^yJ6
As Equation 30 indicates, the equilibrium curve in this case is influenced by the

ideal separation factor So and the pressure ratio r6.
For large pressure ratios, the equilibrium curve simplifies to:
(31)
).n
T pnl Yn
Figure 6.18: Complete mixing.
a = 0.2
---
--- a =0.05 I
0 0.5 xn 1.0
Figure 6.19: Separation characteristics (complete mixing).
Membrane Column
The membrane column can be an alternative to membrane cascades.12-14
The membrane column consequently utilizes the countercurrent flow principle
(Figure 6.20) and, for this reason, is restricted to cases where concentration pol-
arization is negligible. (It has been suggested that the membrane-column concept
be applied to the separation of liquid mixtures. However, this concept must fail
here-at least for asymmetric membranes because of the concentration polariza-
tion within the support layer.) Membrane columns utilize the maximal driving
forces because of the countercurrent flow at both sides of the membrane. The
preferably permeating component is depleted on the high-pressure-side in the
direction of the flow. The preferably permeating component can be withdrawn
at the top of the column, the slower permeating component of the mixture at
the bottom. At present, all modules for membrane columns described in the lit-
erature employ hollow fiber membranes.
Top product
(Permeate)
4
i Enriching section
I
I
I
Feed
I
!
I Stripping section
I
I
I
Bottom product
(Retentate)
Figure 6.20: Membrane column.
The design equations for a membrane column are:
(1) the transport equation for material transport in the membrane

(2) mass, material and energy balances (pressure losses), formulated
for the differential element of the membrane channel.
Material transport in the membrane:

.
(32) n. = Qi- (Xi PI - Yi P2)
1
Mass (molar) balance (M = constant) combined with Equation 32:
.
Material balance:
d’x. * dxi -Ind RT Qi

(34) --$ + =+ x.
P2
--y.
1
dt PlDijd ==D nd2 C Pl l
ij
- xi (xi-yi !5 (,_pz,
i Qi Pl I
For a membrane column, it is essential that axial backmixing by diffusion
(dispersion) be negligible compared to convective transport. In this case, Equa-
tion 34 is reduced to:
d
(35) ~ (xi ~) = TTd Qi (xi P, - yi P2)
Combining Equation 35 with Equation 32 (transport in the membrane) leads to:
dx . 77 d QiPl
(36) -a-$ =
9
Energy-balance: For the calculation of pressure losses, it must be taken

into account-besides the compressibility of gases-that hollow fibers are elasti-
cally deformed (widened) by the higher pressure inside the fiber. According to
Hwang and Thorman,” in the case of a deforming fiber, the pressure losses can
be calculated by:
lbK, nh RI' 8 Re, 2

(37) 2 _ --
.I (t Re;d)-
nd
Pl
P”,,.d z PV d
with Re, =
: rl ; KeZ Iz2 2 rl
= 8 (1+0.75*Ree - 0.0107 Rei + 0.0125 Re:)

Kt
K2 = 1 l 0.056 Rew - O.OlSS*Re~
The set of Equations 32 f 37 has to be solved numerically. For the special case
of total reflux, however, an analytical solution for the concentration profile along
the membrane column can be derived if the pressure losses are neglected. In the
case of total reflux, for every section of the column (Figure 6.21):
(38) x. = y.
1 1
is valid.
ki
Figure 6.21: Membrane column, material balance around a fiber end (total
reflux).
With the boundary conditions:
(39) z = 0: xi = xi a ; i = ka
Equation 36 can be integrated:

l+Oj/Qi Qj /Oi
h x, l-Oj/Qi
&t I
l-Xi, l-Qj/Ql
a lo. -AT Xi l-xi,

(40) z=- -1
Xia l-Xi
nd Qj p,(l-2) 1
-.- 1
l-Oj/Qi
Qj/Qi
l-Oj/Oi
Xia
+1_xla 1
( \-_;;
Xi
--Xia
)
l-Xi,
l-Xi
I-d,,Qi
-1 11
(41) ti = ti, (2)
The separation process is determined (among other parameters) by the re-

flux ratio. Similar to fractionating by distillation, the calculation with total re-
flux leads to the minimal necessary column length. For a given column length,
on the other hand, Equations 40 and 41 indicate the maximum possible purities
of top and bottom product. For these reasons, a discussion of Equations 40 and
41 is useful for a first estimate of the column length and the operating conditions.
Column length depends on the recycled molar flow rate f&. Figure 6.22
indicates that a specific molar flow rate (molar flow per fiber) as low as possible
should be chosen. The lower limit is set by the condition that diffusive axial
backmixing must be negligible with respect to the convective mass transport. On
the other hand, the volume-specific yield (yield per unit-volume of fiber bundle)
should be high.
10
0.1 1.0 21) m
Fiber (Column) length L
Figure 6.22: Membrane column, influence of reflux (recycled molar flow rate)
on column length.
In general, the optimization of a membrane column must consider molar flow

rate &, reflux ratio and the pressure losses. The mutual dependencies, which
have been only shortly discussed here, are the reason why hollow fiber modules
for the membrane column are relatively long (L/d,-20,000) compared to hole
low fiber modules for the separation of liquid mixtures.
Figure 6.23 shows the concentration profile of a membrane column as cal-
culated by Equations 40 and 41 (total reflux). The concentration profile is very
similar to the profile calculated by a numerical solution of the set of Equations
37 for the separation of C0JN2 (Figure 6.24), taking into account the pressure
losses and a finite reflux ratio. Figures 6.23 and 6.24. clearly indicate that a
membrane column is especially effective in the stripping section. For the enrich-
ing section, especially for high concentrations of the preferably permeating com-
ponent, the necessary membrane area and the reflux ratio (pressure losses) are
increasing rapidly. It should be emphasized that a membrane column represents
a multi-stage process just as a reflux cascade. The question, whether a membrane
column or a reflux cascade is superior cannot be answered in general but must
be evaluated for every individual case. It seems doubtful, however, whether
multistage membrane processes will be often employed. According to our opin-
ion”, only one stage or, at the most, two-stage cascades seem to be competitive
compared to proven processes like pressure swing adsorption, absorption, or
low-temperature distillation.
I I
0.8 ------ ---
Pl = 10 bar
0 0.2 04 0.6 0.8 m 1.0

Fiber (Co(umnI length L
Figure 6.23: Calculated concentration profile of a membrane column. total

reflux.
Feed = 52.6 % CO2
0 Experiment
- Theory ( Thorman 1
I I I I
0 1 2 3 4 m
Length
Figure 6.24: Concentration profile of a membrane column, comparison be-

tween theory and experiment.13
PROCESSES
Any application of membrane processes will be ultimately determined with

respect to economic reasons, when compared to proven separation processes
such as distillation. Strictly speaking, membrane processes will not replace con-
ventional processes but, in the main, a combination of membrane processesand
conventional processes will be the optimal solution. This trend can be noticed at
least partly in the following examples.
Seawater Desalination by RO
At present, nearly all seawater desalination plants are distillation plants.

Since low pressure steam can be utilized for heating, all major plants are com-
bined with power plants-the heat input section of the desalination units replac-
ing the condenser in the Rankine cycle. By this combination, the primary en-
ergy consumption of seawater desalination can be kept Iowl which is one of
the reasons that thermal processes will always have their place in seawater de-
salination although the relatively rigid coupling of power and water production
is sometimes disadvantageous.
Modern distillation plants have unit capacities of about 25,000 t/d. The
energy consumption is approximately 95 KWh/to distillate in the form of sat-
urated steam of a pressure of about 2.2 bar and 4 to 5 KWh/to distillate in form
of power for the pump drives. All distillation processes produce a distillate con-
taining only 20 to 30 ppm salts (total dissolved solids). Reverse osmosis (RO)
will be the alternative to distillation. The modules which are successfully em-
ployed are of the “spiral wound” or the “hollow fiber” configuration. Under
favorable conditions, the economic service lifetime of the modules resp. the
membranes is about 5 years and warranties for such a figure are given by manu-
facturers.
RO plants are, however, much more sensitive to insufficient feedwater pre-
treatment than thermal plants. For this reason, it is accepted technology to use
well water from wells drilled at the seashore rather than surface water as in
thermal desalination.
As an example of the present state of the art in RO-seawater desalination,
some data of the Malta facility shall be discussed.” The entire plant, consisting
of 10 trains of hollow fiber modules (BlO Permeators of DuPont), produces
20,000 t/d fresh water at a specific power consumption of 6.3 KWh/to fresh
water. This very low registered figure includes the power consumption for the
feed pretreatment and is a direct consequence of the installed energy recovery
system (Guinard integrated turbo pumps). Figure 6.25 shows the process flow
sheet of the plant. The feed pretreatment is simple: sand-filtration (well water)
followed by a settling tank, acid dosing and cartridge filtration. Feedwater tem-
perature is not a critical factor in this case (in general, the manufacturer limits
the maximum temperature to about 30°C despite the fact that the membrane
itself can withstand higher temperatures). Table 6.1 summarizes the major op
erating data of the plant; Table 6.2 is a breakdown of cost.
In general, the following measures have to be considered for feed pretreat-
ment of RO seawater plants:‘s
l pH-adjustment by acid dosing against carbonate scaling

a chlorine dosing against marine life in the plant
Design and Optimization
Process 375
0 polyphosphate dosing against calcium sulfate scaling

0 cartridge filtration.
Verlical
-‘:
Well,
(15.3)
Feed
Break
Tank
Feed Boosf
Pumps
,---_A (‘+‘I 1 Product Storage
Tank
t5*11
I
- Caurtlc
t -
I
1
I
I
t + t
R.O. SYSTEMS
(10)
BRINE TO SEA
37.000 Cum/d
Figure 6.25: 20,000 cu-m/d seawater R.O. plant, Ghar Lapsi, Malta-process
flow.
Table 6.1: Malta Seawater RO Plant Initial Operating Data
ACTUAL DESIGN
ITEM UNITS DATAa BASIS
FEEDWATER
Total Dissolved Solids 36,500 39,200
Temperature rr;* 17
pressure bar 70 ;;
Recovery % 33 3.5
Silt Density Indexb 2-4 o-3
PRODUCT
Flowrate m'/d 18,600C 20,000
Total Dissolved Solids pp* 380 500
CONSUMABLES
Electricity KWh/m' 6.30 6.12
Sulfuric Acid pp* 6.6 6.6
Caustic Soda pp* 3.2 3.2
Chlorine pp* 1.0 1.0
I-
a 13 March 1983 (400 Operating Hours)
b Before cartridge filters
' 80% of permeators installed
Table 6.2: Production Cost Summary Malta Seawater RO Plant
S/m’ S/Kgal
Electricity 0.43 1.62
Permeators 0.11 0.43
Labor L Overhead 0.05 0.20
Spare Parts 0.04 0.14
Chemicals 0.01 0.03
Filters 0.01 0.04
Total Operating 0.65 2.46
Total Investment
costs RO-plant 1.250.000
Civil works 5.000.000
Total production 1.08 4.07
Furthermore, a flushing and cleaning system is necessary. For membranes

which are sensitive to oxidizing agents, active charcoal filtration and/or sodium
bisulfite dosing is mandatory if chlorine has been used in the pretreatment. Note
that the product of any one-stage RO seawater plant contains about 300 to
500 ppm total dissolved solids, i.e., much more than the distillate from thermal
desalination plants. Higher product purities-for chemical processesor for boilers-
can only be achieved by a (2-stage) RO cascade. The second stage can be equipped
with a different type of membrane (high-flux) than the first stage and certainly
must be operated at a different transmembrane pressure difference. The module
arrangement in each stage, i.e., the number of modules in series and/or in par-
allel is determined for a given type of module by the desired yield of the stage
and a necessity to maintain optimal flow conditions in the modules (concen-
tration-polarization). In general, these requirements lead to a so-called “tapered
design” (see Figure 6.13) In any case, the designer faces on optimizing problem
which can be solved numerically by a relatively simple computer program based
on the general flow sheet indicated in Figure 6.26. One result of numerical cal-
culations which seems to be generally valid is that recycling pumps (ZP in Fig
ure 6.26) for the improvement of the flow conditions in the modules and/or
pumps between the module-banks of a single stage (DEP in Figure 6.26) are
not economical. More details about seawater desalination by RO can be found
in the literature, especially in the proceedings of the IDEA conferences and of
the symposia “Fresh water from the Sea”,1912o
Total Desalination of Brackish Water

When desalting brackish water far inland, the discharge of the concentrated
brine may pose a serious problem. For the case where neither deep-well-injection
nor a discharge into a river or canal is possible, a hybrid plant consisting of a re-
verse osmosis process as a first stage and a special crystallization process as a sec-
ond stage seems to be feasible (Figure 6.27).
SEAWATER
CONCENTRATE
HDl
m. STAGE
FRESH WATER
Figure 6.26: General flow sheet of an m--stage RO cascade.
1 Cry s’ar’ion]
Solid waste
Figure 6.27: Total desalination of brackish water by a combination of RO and

crystallization.
Studies on three real brackish waters of different sources indicated that for
the RO stage the optimum pressure will be between 40 and 50 bar. If modules
of the spiral wound type are employed, for example ROGA 4160 HR. the
Reynolds number, Re = h.v&,, at the inlet should be about 80. Although the
Reynolds number has a significant influence on the plant structure, i.e., on the
ratio of modules in series and in parallel, its influence on the minimum specific
cost of permeate is nil in a wide range (Figure 6.28).‘l Furthermore, the calcula-
tions clearly indicate that internal recirculations and/or repressurizing pumps be-
tween the module banks of a single stage will not pay off (see Figure 6.26). For
spiral wound modules (Figure 6.29) as well as for hollow fiber modules (Figure
6.30). the lowest costs are obtained without recirculation (z = 0) and without
repressurizing pumps (in Figures 6.29 and 6.30). Costs due to frictional losses
and additional pumps are higher than the possible gain in permeate-flux by
improved flow-conditions.
The second stage of the hybrid-process is essentially a crystallization capable
of crystallizing all solids dissolved in the feed resp. the brine of the RO stage.
Parts of these substances tend to heavy scaling and, therefore, the choice of the
crystallization equipment is limited. A proven solution is, of course, the agitated
thin film evaporator. However, this type of equipment must lead to high specific
treatment costs for two reasons:
(1) An agitated thin-film evaporator is costly and limited iv: size.

(2) Because of its high capital cost, it is in general out of the WeStiOn
to combine agitated thin film evaporators to an energy SaVing mul-
tiple effect evaporation process.
5 IO 15
Number of modules in series
Figure 6.28: Influence of flow conditions in the modules on the specific cost
of RO (tapered arrangement).
without
repressurizing
--- with I
Roga 4160
Re = 120 I
Module design : 4 /2/l
_-
I LO 60 bar
Pressure Ap
Figure 6.29: The influence of brine recirculation within each bank of modules
(r$++Jti~ = z > 0) on specific cost of RO (spiral-wound).
1.;
DrJ
t
1.1
f;;
00
IS
0.9 I
- without
repressurizing
--- with I
Permasep B9
Re = 50000
Module design: Z/l/O
0.8
20 40 bar 60
Pressure Ap
Figure 6.30: Influence of brine recirculation within each bank of modules
(mRec/mF = z > 0) on specific cost of RO (hollow fiber).
Low treatment costs for the brine can only be expected if the heat transfer sur-
faces can be designed as tube bundles and if energy saving concepts like MSF,
ME or VC can be applied.
Promising crystallization processes like the seeding process and a modified
MSF process, employing an inexpensive oil as heat carrier, are discussed in Ref-
erence 22.
One of the most important variables for the hybrid process is the concentra-
tion of the RO brine. In principle, the specific treatment costs of each stage
must increase with increasing concentration of the RO brine. For the RO stage,
this is mainly due to the rising osmotic pressure. For the crystallization step not
concentration per se, but the fact that a rising brine concentration in the RO
stage is equivalent to a decrease in feed flow for the crystallization step, results
in smaller but specifically more costly crystallization units. However, because
the specific treatment cost of the crystallization is in any case higher by an order
of magnitude than the specific treatment cost of the RO step, it is easily under-
stood that the optimum of the hybrid plant is determined by the maximum con-
centration which can be achieved with the RO modules! This result is demon-
strated by Figure 6.31 comparing specific costs of different process combina-
tions.
Capacity RO
ATF: Agitated Ihin Eilm Evaporator

VTE: Vertical Tube Evaporator
MSF: Multi Stage Elash
Figure 6.31: Water cost of alternative hybrid plants,

In any case, the specific costs of fresh water for a 2-stage hybrid process
will be 3 or even 4 times higher than for a single-stage RO process. This is a di-
rect consequence of the specific costs of the second stage, regardless of whether
a VTE-seeding process, a modified MSF-process or vapor compression is consid-
ered. (VTE = Vertical Tube Evaporator, here in combination with a seeding tech-
nique, MSF = Multi-Stage-Flash, ATFE/VC = Agitated-Thin-Film-Evaporator in
combination with Vapor Compression). Because of these high treatment costs,
the chances for such a hybrid process are very limited with respect to brackish
water desalination. The situation seems to be different for wastewater treatment
and it is not surprising, that a process as described has been realized for the treat-
ment of the effluents of a power plant (San Juan, New Mexico, USA).
The process is designed to handle 10.000 m3/d wastewater from the flue
gas scrubber and the cooling towers (Figure 6.32). The mechanically pretreated
effluents are fed to a RO unit, designed for a recovery rate of 80%. The concen-
trate of the RO unit is fed to the crystallization stage, designed as vertical tube
evaporator combined with vapor compression. Scaling in the evaporator is pre-
vented by the application of a (calcium-carbonate) seeding technique.
Cleonlng Feed waler COllVENTIOt4AL

of flue gas treatment PRETREATMENT EVAPORATION
llue dust removal
Calcium- Addatives vopour -
__..^^_a^ I I
__---L___,
I
Permeate
Figure 6.32: Treatment of power-plant wastewater; combination of RO and

crystallization.
Automotive Industry-Combination of UF, Cross Flow Filtration and Evaporation

for Recycling of Detergents and Process Water
The final step in the production of automotive parts such as pistons, axles,
etc. is usually a rinse. The spent wash water of such a process contains, in addi-
tion to the detergents, cutting oil and suspended matter and it is impossible to
discharge it into the canalization without further treatment. Fractionating the
wastewater into a concentrate containing the oil and the suspended particles and
into a premeate containing practically no oil is possible by UF.
The detergent concentration of the permeate and, consequently, its biologi-

cal oxygen demand (BOD) is high, leading to further treatment costs if discharged
into the canalization. For two obvious reasons, the question arises whether such
a permeate could be recycled:
(1) a further treatment or further treatment costs would be obsolete and

(2) the recycling would lead to substantial savings in process water and
in detergents.
The savings in detergents are of prime importance and it must be the aim of
process development to maximize the detergent concentration in the permeate
without increasing the oil-concentration to an intolerable amount.
In on-line experiments with a plant effluent containing 2% detergents,
1% oil and 0.8% suspended solids (metal, dirt, etc.), tubular modules equipped
with UF membranes of different pore-sizes and a microfiltration membrane have
been tested. (The experiments were carried out in the gear and axle production
of Daimler Benz AG.) As expected, the MF membrane produced the highest flux
for detergents. Because of th relatively low retention capacity for oil, however,
the use of MF membranes is limited to a maximum oil concentration of about
10% in the concentrate. In this case, the oil concentration of the permeate is
about 50 mg/S?, contrary to the UF membranes, where the oil concentration
of the permeate is independent of the oil concentration of the concentrate up
to the phase inversion concentration of about 41%.
For ultrafiltration as well as microfiltration the permeate flux is controlled
by a gel layer-not by the membrane itself. This must be concluded from Figure
6.33 because the permeate flux for steady state conditions is almost exactly the
same for the three tested membranes (for comparison, the data of Figure 6.33
have to be corrected for the same temperature and the same flow conditions).
@a Membrane
XT, 30m/s, 33 bnr
0 10 20 30 50 60 min
Figure 6.33: Gel-layer controlled flux in UF and MF (MF-membranes with and

without backflushing).
Quite often, the formation of a gel layer is reversible; the gel layer can be
removed by cleaning. In every case, periodic cleaning and a higher average flux
have to be weighed against the time for cleaning. Microfiltration is advantageous
compared to UF because microfiltration membranes are available in the form of
tubes withstanding outer or inner pressure of several bars which can be easily
cleaned by reversing the flux for a short time. Despite the advantages with re-
spect to flux and yield of detergents, microfiltration alone will not be a solution
for the described effluent-treatment problem since the retention capacity for
oil at higher feed concentrations is too low. The optimal process is a combina-
tion of microfiltration and ultrafiltration according to Figure 6.34.24
MICROFILTRATION ULTRAFILTRATION
11.6 m3/d 1.16m3/d
FEED F F=2,Om2,Ap=2bor .F=1.6m2,Ap=3.3bor
1 % Oil 10% Oil
T=38OC, Y =S.Sm/s T=36*C, Y =3m/s
TO COMBUSTION 0.12 t-n31 d

1 EVAPORATION 1-1
OR REFINERY 90 % Oil
Figure 6.34: Flow sheet for a combination of micro- and ultrafiltration for the
treatment of oily wastewater.
The recovery of detergents of such a process has been determined to an av-

erage of 85%. It should be noted, that the permeabilities of the individual com-
ponents of a detergent differ and that the upgrading of the recycled permeate
must be based on the selectivity of the process.
The limiting oil-concentration for the first stage, microfiltration, is deter-
mined by the tolerable concentration of oil in the recycled product. Since the
oil retention capacity of the ultrafiltration stage is independent of feed concen-
tration, the final concentration of the retentate is solely determined by the
phase inversion “oil/water -+ water/oil”. At this concentration, the overall water
recovery of the process is above 97.5%. However, the retentate of the process
still contains too much water if incineration or refining is considered. Therefore,
an evaporation step must be included in the process. Almost certainly evapora-
tion will not be economical for the small capacities indicated in Figure 6.34 and
a central evaporation station for the tentates from several production lines
should be considered.
Galvanic Industry-Treatment of Effluents

In the galvanic industry, a treatment of the effluents is mandatory. A
process which is not only capable of water treatment but, at the same time, of
reclaiming metals which would otherwise be lost would be most favorable. RO
is such a process. In the U.S., more than 200 RO plants are installed in the gal-
vanic industry and, according to reported data, these plants are very profitable
because of the reclaimed metals.
Figure 6.35 is a flow sheet of a galvanic production line completed by RO.
The modules are of the spiral wound type and the membrane material is cellu-
lose acetate in this case.25
Evoporat~on losses
Ltquid 0.04llmm l0.3at/mm
Orogglng losses 0,04llmIn
Delonlrlng wa!er
I I
0.3allm8fl
Golvoniz!ng balh
2aoooomgII
9.Sml O,oLl/mln
I c 1 1 * I_
1027mgll - 407mgll 37mgll
,,1?04
Ilmin
LOBrngll
1703 ltmin 11,361Imm

9663mgA 14300mgll
Reverse 0srnos15 ’ Reverse 05nlos~*
5.6aIlmln
575mg11
5,6alImm I
zadoomg/I
Figure 6.35: Flow sheet of nickel plating, inclbding RO for wastewater treatment.
By recycling of the permeate into the first stage of the rinsing cascade and
recycling the concentrate into the galvanic bath, the recycling is complete; the
evaporation losses of the plant are balanced by the deionized water necessary in
the last stage of the rinsing cascade. Depending on the permeate concentration
which, in turn, depends on membrane retention and on the required recovery
rate of the RO unit, the permeate must be added to the first or second stage of
the rinsing cascade. Table 6.3 shows the (average) composition of the galvanizing
bath; Figtire 6.36 shows flux and selectivity (retention rate) of the modules.
Table 6.3: Composition of Watts-nickel bath
concentration
total nickel ratio 82 g/l
nickel sulfate 6 H20 255 g/l
nickel chloride 6 HZ0 105 g/l
boric acid 45 g/l
brightening agent not analyzed
pH- value 4.3
temperature 60 =C
80
60
50
0 200 LOO 600 800 1000 1200 1400 h
Time
I
-ii
600
200 LOO 600 800 1000 12

Time
Figure 6.36: Treatment of nickel-plating effluents by RO, module flux and

selectivity.
Figure 6.36 indicates the high retention capacity of the membranes for nickel
and solubles whereas the retention for boric acid is not quite satisfactory (no
explanation could be found for the concentration fluctuations in time).
Payback time of the RO units is given as 8.1 years for single shift operation,
2.4 years for a two shift operation and 1.4 years for a three shift operation.
An analysis of the RO units installed in the galvanic industry demonstrates
that presently RO units are mainly installed in the nickel plating industry for
two reasons:
(1) the pH value of Watts nickel baths is between 3 and 5 permitting

the installation of cellulose acetate membranes (cellulose acetate,
still being a very common membrane material). It should be men-
tioned that more recently synthetic membranes have been devel-
oped (i.e., the PA 300 of UOP, San Diego, the NS 100 of Abcor
Inc. Wilmington or the PBIL of Teijin Ltd., Tokyo) which proved
to be resistant to chromium acid, copper cyanide and zinc cy-
anide baths at extreme pH values.
(2) the temperature of the Watts nickel bath is 50” to 70%. At this
temperature, the evaporation losses of the bath are such that no
additional evaporators for further concentration of the RO con-
centrate are necessary.
Gas Permeation
With few exceptions, gas permeation on a technical scale employs mem-
branes of the sorption diffusion type. In this case, the flux of a permeating com-
ponent is proportional to the difference of the partial pressures at both sides of
the membrane.
Ji = Qi (PlXi - P2 Yi)
at least as long as the conditions of the gasesare well above the critical point. On
a commercial basis, gas separation started with the recovery of Hz from the bleed
of (high pressure) synthesis loops, employing in most cases a composite mem-
brane “silicon/polysulfone” in the form of hollow fibers.
Asymmetric phase-inversion membranes like the membranes employed in
reverse osmosis are difficult to prepare as gas permeation is much more sensitive
to micropores than RO due to the much higher diffusion coefficients of gases.
For the same reason, the composite membrane differs from RO composite
membranes: in gas permeation, the top layer of the asymmetric support struc-
ture is responsible for the separation while it is the sole duty of the coating to
plug the micropores. Consequently, the material of the coating chosen (silicone)
has a high permeability but a low selectivity while the membrane material (poly-
sulfone) has a high selectivity (and a much lower permeability).
Figure 6.37 shows the flow sheet for the recovery of Hz from the bleed of
the synthesis loop of ammonia plants. In ammonia synthesis, a recycling of the
unreacted components is mandatory because of its low equilibrium conversion
rate. In all technical processes, however, a bleed is necessary because otherwise
the concentration of impurities would increase to an intolerable level. Normally,
the bleed is utilized for heating purposes in the reforming stage. Here, the bleed
is fed to a conventional separation unit first-a scrubber for recovering ammonia.
Behind the scrubber, the modules for the recovery of Hz are arranged in a “one-
stage-two unit” form. The first unit, consisting of 8 hollow fiber modules (total
feed capacity 3,800 Nm3/h) is operated with a transmembrane pressure differ-
ence of 60 bars, the permeate leaving at a pressure of about 70 bars. At this pres-
sure, the permeate can be fed to the second stage of the synthesis feed compres-
sor.
05 9Sbor
Retentate 4 ’
Water
,
I
I
i 25bar
I
-1,
Feed
AMMONIA -
SYNTHESIS
I
+ Ammonia
Figure 6.37: Flow sheet for the recovery of Hz from the synthesis loop of an
ammonia plant.
The retentate of the first unit is fed to the second unit (and for this reason,
it cannot be considered a two stage cascade), the permeate leaving at 25 bars as
an additional feed to the first stage of the compressor. The retentate is utilized
for heating pur?osesz6
Since flux and selectivity increase with increasing transmembrane pressure
difference and the modules can tolerate pressure differences of about 166 bars,
the reader might question why the first unit is operated with a transmembrane
Pressure difference of only 60 bars. The reason is that the Hz recovery system has
been added to an existing plant and that the first stage of the synthesis compres-
sor would not accept the permeate flux of both units.
Figure 6.38 shows another example of Hz recovery; here from the tail gas of
a high pressure synthesis (UOP Butamer process). In this case, the tail gas leaving
the conventional fractionating system contains about 70% HZ. The membrane
unit placed behind the fractionating system recovers about 90% with a Purity of
>96%.27 This process seems to be interesting for two reasons:
(I) spiral-wound modules are employed with dry asymmetric phase in-
version (cellulose-acetate) membranes.
(2) the feed contains small amounts of HCI.
ISOEiJTANE
PRODUCT GAS
Figure6.38: Hydrogen recovery from the tail gas of the UOP Butamer proc-
27
ess.
Figure 6.3927 shows a two-stage cascade for the purification and dehydra-
tion of sour gases, mainly removing CO2 and H2S. Again, spiral wound modules
with asymmetric cellulose acetate membranes are employed. It should be noted,
that in this case as in all other cases discussed here, no compressors had to be
installed. This is the main reason why these applications show excellent payback
times.
TO AMINE UNIT -
FEED @
8
N
TO SULFUR RECOVERY
UNIT
VOLUME FLUX
I Nm3/h I 17110 29830 12625 16872 12712 L130
MOLAR FLUX
[kmol /h I 763 1331 563 753 566 10L
PRESSURE
lbarl 66,5 66,5 65,5 18,2 17,2 1,35
Figure 6.39: Purification of sour gas by a two-stage stripping cascade.27
Pervaporation
Pervaporation differs from all other membrane processes because of the
phase change of the permeate. Masstransport across the membrane is not achieved
by elevated pressures at the feed side as in RO and in gas permeation but by low-
ering the activity of the permeating components at the permeate side. This can
be achieved by applying either a vacuum at the permeate side or by sweeping
the permeate side with a carrier like air or Hz0 vapor (Figure 6.40). Phase
change occurs because the partial pressure of the permeating components is
lower than the corresponding saturation pressure.
Figure 6.41 shows that the selectivity decreases with increasing pressure at
the permeate side. For a ratio of p/p; = 1 finally, the separation characteristic
is identical to the thermodynamic equilibrium curve (for 19= constant). Pervapo-
ration will always be a process which is relatively expensive compared to other
membrane processes for two reasons:
(1) The modules must be designed for a low pressure drop at the per-
meate side despite the increasing volume of the permeate due to
the phase change since the principle of pervaporation is very sensi-
tive to such pressure losses.
(2) The process requires heat transfer surfaces because the heat of
evaporation necessary for the phase change of the permeate must
be supplied to the process (and must be rejected by condensation
of the permeate).
The necessary heat-flux
;I = Z Ji (Ahvi + c A91
piv
is usually drawn from the feed and will inevitably lead to temperature gradients-
orthogonally to the membrane as well as in the dirction of flow. Especially in
pervaporation of water mixtures with water as the preferably permeating com-
ponent, heat transfer between the bulk of the liquid and the membrane surface
(permeate-side) can become the rate-controlling step.
Figure 6.42 shows the temperature drop orthogonally to the membrane as
a function of heat flux for water/isopropanol flowing in a rectangular channel
(laminar flow, Reh = 498, dh = 4 mm). According to Figure 6.42, the correlation
between measurements and calculations (straight line) based on
is fairly good. ” Quite often the heat-flux is small and, therefore, the tempera-
ture drop orthogonally to the membrane can be neglected. In any case, the tem-
perature decrease in the direction of flow has to be taken into account.
Figure 6.43 shows, for the pervaporation of benzene/cyclohexane and for
symmetric PE-membranes, the temperature decrease along the membrane. The
temperature distribution has been calculated numerically assuming laminar flow
in a rectangular channel with membranes at only one side. Due to the laminar
flow, heat transfer orthogonally to the membrane is only achieved by heat con-
duction and, for this reason, the channel height should be small. The dotted lines
in Figure 6.43 are valid for external heating of the channel wall opposite to the
membrane. Such a design is expensive. In most cases, it will be the far better so-
lution to achieve a quasi-isothermal operation by a combination of modules and
heat exchangers connected in series.
5-
‘C
I-
3- --
Membrane Mcmbrone Srm I
2-
l-
o’
I
I
I
I
ICAl
I
1
1 0
I
I
I
1
I
1 1500 2000 2500 3000 W/m2

Heat FLUX 4
Figure 6.42: Temperature difference orthogonally to the membrane between

bulk of feed and permeate.
I I /
BenzenelCyclohexane
PE -Membranes
Rear= SO,wi,=O.S,3,=L
- ile, :o
---
% :180Wlm2
-q ’ ’ ’ I I I I I I -
I I II
10’ lo2 mm IO3
Channel Length L
Figure 6.43: Temperature distribution of the feed along a membrane channel.

The major field of application for pervaporation will be either the separa-
tion of organic components with almost identical boiling characteristics or azeo-
tropic mixtures. In most cases, the desired product quality cannot be achieved in
a single step-either a combination of different processes or a “multi-stage” proc-
ess will be necessary. In this case, a reflux-cascade has to be designed.
Figure 6.44 shows some results of a numerical optimization of pervapo-
ration cascades.32 Assumptions: (I) the selectivity is constant and equal for all
stages; (2) the fluxes are small enough to neglect concentration polarization.
10
6,
Number of Stages
a 10 20
Selectlvlly Sn
Figure 6.44: Optimisation of a reflux cascade for pervaporation: influence of

selectivity on total permeate flow and number of stages.
Mostly, cascades are designed for either constant reflux, or constant cut
rate or as a so-called “ideal cascade”. Here, a different optimization procedure
has been chosen: with respect to costs, the permeate flows are of prime impor-
tance and, therefore, the calculation was aimed at minimizing the sum of all per-
meate flows in the cascade.
Figure 6.44 clearly indicates that for a decreasing selectivity of the module
the specific process costs will increase for two reasons:
(1) the number of stages increases and

(2) the membrane area increases because the internal permeate (reflux)
flows increase.
There already exist applications of pervaporation on a technical scale. A

hybrid process (Figure 6.45) for the production of pure alcohol, combining
distillation and pervaporation, proved superior to the conventional approach
(extractive distillation) with respect to specific energy consumption (1 kg/kg
ing to Figure 6.45, the alcohol concentration is increased from about 80% to
99.5% by only one pervaporation stage.33 It should be noticed, that for this
range of concentrations membranes for which water is the preferable permeating
component must be used.
1 r--l I I
Ethanol
__ _ 99,a %
I
r
I I
Vacuum
waste Permeate
Feed
Figure 6.45: Hybrid-plant for production of pure alcohol.
Although the high separation potential of permeation has been demonstrated

for a number of systems in laboratory experiments,30 pervaporation seems to be
economical only in cases where high product purities are required-and in com-
bination with a conventional separation process. This statement is based at pres-
ent only on a detailed analysis of the separation of benzene/cyclohexane and
corroborated by the abovementioned production of pure alcohol, but there can
be no doubt that the results are valid for many other systems as well.
The fractionating of a 50% benzene-cyclohexane mixture into products of
98% purity is normally achieved by extractive distillation with furfural as a car-
rier. If pervaporation is to be employed instead, at best a three stage cascade
with the refluxes indicated in Figure 6.46 is needed, where each stage consists
of a combination of membrane-modules and heat exchangers. In a numerical
calculation, the temperature decrease “along” each module is a free parameter.
In Figure 6.46, this temperature decrease is set equal for each module (7.5 K).
This leads directly to the necessary number of heat exchangers and membrane
units per stage and to the necessary heat transfer area.
The flow-sheet indicates a major difficulty of the process: if the membrane
selectivity is high, fluxes will vary markedly with the feed concentration of the
preferably permeating component. Permeate fluxes might differ by a factor of

10 between the first and the last stage if the same membrane is employed
throughout the cascade. Compared to extractive distillation, a pervaporation
cascade will not be competitive (Table 6.4); investment costs as well as operating
costs are much higher. Especially noted should be the high costs for condensa-
tion-a direct consequence of the low pressure (150 mbar) at the permeate side.
(Entrance temperature of the cooling water was assumed to be 2O’C.)
The situation is different if product purities of about 99.5% are required. In
this case, the energy consumption of the extractive distillation is very high be-
cause of the high reflux ratio (in the extractive distillation tower, the effective-
ness of the carrier is rapidly decreasing in the section above the feed tray for the
carrier and as a consequence, benzene will “enrich” in this section).
COOLING WATER -‘t’T
HEATING
Figure 6.46: Flow sheet of an optimized cascade for the separation of a 50%
benzene-cyclohexane mixture, product quality 98% benzene and cyclohexane.
(a19 = 7.5 K per membrane unit).
A hybrid process according to Figure 6.47 with a one-state pervaporation

for final purification of the top product of, the extractive distillation shows a
cost advantage of 20%, compared to pure extractive distillation, even for very
conservative assumptions. The calculations are based on the following assump-
tions:
(1) Capacity NF = 40 kmol/h

(2) Equimolar feed of benzene (1) - cyclohexane (2)
(3) Product purities xzl A 0.992 and x12 = 0.995

(4) Effective thickness of the membrane = 8 pm
The optimal feed concentration of the pervaporation unit depends on car-

rier flow rate, reflux ratio and number of theoretical trays of the extractive dis-
tillation. Retentate concentration and cut rate of the pervaporation stage follow
from the requested product quality xl1 + x3, GO.008. For the design of the per-
vaporation stage, the worst case has been assumed that only benzene and no fur-
fural (3) will pervaporate. The major factor for the cost reduction is the much
lower energy consumption of the hybrid process of 1.18 t/h heating steam
against 1.7 t/h for the conventional process.
Table 6.4: Extractive Distillation vs. Pervaporation-Cost Analysis
DE1
spec. costs
3::::::::1.,,
J sieve-tray column 6,11 I __

2
u module -_ 4.85
e, evaporator/heat exchanger 4,43 2,67
2
condenser 1.41 6.33
5
: Pomps 0.45 0,72
z piping 1,86 2.19
membranes (Dbl 100/m* 1 __ 43,65

z
.rl steam 18.81 14.44
+
2:: cooling water 1 ,38 12.69
E.;: power 0,09 0.04
ou
total costs 34.54 87,58
Assumptions
1) membrane service lifetime 20 months
2) cost of piping = 15 % of cost of main items
3) linear amortization within 5 years,
10 % interest
4) Pervaporation: $)u = 75 “C
9v = 3 K per membrane unit (optimal A81
= 8 urn .
%i
LIW
reflux ratio in stage 1: i = 0.75
= 150 mbar lo
p2
extractive distillation: reflux ratio
column I: vI = 2.5, column II: VII = 1
QC
l
Benzene
x,rr = 0,995
=:
Figure 6.47: Flowsheet of a hybrid-process “extractive distillation-pervapo-

ration” for the production of 99.2% cyclohexane.
According to Table 6.5, a further reduction of the specific separation costs

is feasible if the effective membrane thickness can be reduced, for example, by
employing “optimized” asymmetric membranes.31 It should be noted, that the
optimal feed concentration of the pervaporation unit is not only determined by
the design of the extractive distillation but by the effective membrane thickness
(or specific membrane costs) as well. With increasing permeate fluxes, or lower
specific membrane costs, the optimal feed concentration is shifted to higher
benzene concentrations.
Table 6.5: Hybrid-Process, Influence of Membrane Thickness on Process Costs

and Optimal Intermediate Concentration xlDl of the Hybrid Process
Feed Concentration Product Concentration . bMa L costs [mm 1

Xla X2a x3. xlW x2w xsw [urn) PV +' Dist. Total
0.0062 0,991l 0.0027 0,OOSZ 0,992O 0.0028 8 10.55 63.96 74.51
0.0071 0,9899 0.0030 0,0048 0,992O 0,0032 4 Il.41 53,39 64,80
0.0100 0.9865 0.0035 0,0039 0.9920 0,0041 2 14.25 41.18 55.43
Extractive Distillation 0.0055 0.9920 0.0025 I 93.65
+’ Reference-Stream: Retentate + Distillate of Column II

REFERENCES
1. Sherwood, T.K., Pigford, R.L., Wilke, C.R., Mass Transfer, McGraw-Hill

Kogakusha, Ltd., Tokyo, 167/169 (1975).
2. Blaisdell, C.T., Kammermeyer, K.,Chem. Eng. Sci., 28, 1249 (1973).
3. Gill, W.N. and Bansal, B., AIChE J., 19, 823/831 (1973).
4. Hwang, S.T. and Thorman, J.M., AlChE J., 26,558/566 (1980).
5. Kammermeyer, K., Chem.-/ng,-Tech., 48, 672/675 (1976).
6. Weller, S. and Steiner, W.A., J. Appl. Phys., 21, S. 279 (1950).
7. Naylor, R.W. and Backer, P.O., AlChE J., 1, S. 95 (1955).
8. Blaisdell, C.T. and Kammermeyer, K., Chem. Eng. Sci., 28, S. 1249 (1973).
9. Breuer, M.E. and Kammermeyer, K., Seperation Science, 2, 319 (1967).
10. Hwang, S.-T. and Kammermeyer, K., “Membrane in Separations”, Tech-
niques of Chemistry, Volume VII, John Wiley & Sons (1965).
11. Walawender, W.P. and Stern, S.A., Separation Science, 7, 5, S. 553-584
(1972).
12. Thorman, J.M., Engineering Aspects of Capillary Gas Permeators and the
Continuous Membrane Column., Dissertation, University of Iowa (1979).
13. Hwang, S.T. and Thorman, T.M., AlChE J., 26,4, 5581566 (1980).
14. Hwang, S.T. and Ghalchi, S., J. Membr. Sci., 11, 187/198 (1982).
15. Rautenbach, R. and Dahm, W!, Chem. Eng. Technol. 10,256-261 (1987).
16. Rautenbach, R. and Arzt, B., The Arabian Journal for Science and Engi-
neering, Vol. 8, No. 3 (1983).
17. Andrews, W.T. and Berman, R.A., “The Malta Seawater RO Facility”
presented at the First World Congress on Desalination and Water Re-
Use, Florence (Italy) 23-27 (May 1983).
18. GVC, Seawater Desalination-Water Pretreatment and Conditioning, Edi-
tor: VDI-Gesellschaft Verfahrenstechnik und Chemie-ingenieurwesen,
Diisseldorf (FRG).
19. First World Congress on Desalination and Water Reuse, Florence (Italy)
(May 1983).
20. Proceedings of the 7th International Symposium on Fresh Water from the
Sea, Amsterdam, 23-26 (September 1980).
21. Henne, K.-H., Ph.D. Thesis, RWTH Aachen (1981).
22. Rautenbach, R. and Henne, K.-H., “Total Desalination of Brackish Water,”
Proceedings of the 7th International Symposium on Fresh Water from
the Sea, Amsterdam, Vol. 1, 163-171 (1980).
23. Paul, H., Schock, G. and Rautenbach, R., Maschinenmarkr, 89, S. 1785-
1988 (1983).
24. Rautenbach, R., Schock, G. and Paul, H., Ferre, Seifen, Ansrrichm., 86, 1,
l-46 (1984).
25. McNulty, K.J. and Goldsmith, A.Z., Gollan, US-NTIS-Rep. PB 266919
(Feb. 1977).
26. Monsanto-Communications, Symposium in Scheveningen (NL) 24.03.
(1982).
27. Schell, W.J. and Houston, C.D., Use of Spiral-Wound Gas Permearors for
Purification and Recovery ( 1982).
28. Heisler, E.G., Science, 124, 77/78 (1956).
29. Binning, R.C. and James, F.E., Petroleum Refiner, 37, 5, 214-216 (1958).
30. Aptel, P., Cuny, J., Challard, N. and Neel, J., J. Membr. Sci., 1, 271-287
(1976).
31. Rautenbach, R. and Albrecht, R., On the Behavior of Asymmetric Mem-
branes in Pervaporation, European Workshop on Pervaporation, Nancy,
21-22.9 (1982).
32. Albrecht, R., Pervaporation-BeitrSge zur Verfahrensentwicklung, Ph.D.
Thesis, RWTH Aachen (FRG) (1983).
33. Ballweg, A.H., Briischke, H.A., Schneider, W., Tusel, G.F., et. al., Per-
vaporation Membranes-An Economical Method to Replace Conventional
Distillation and Rectification Columns in Ethanol Distilleries, 5th Int.
Symp. on Alcohol Fuel Techn., Auckland, New Zealand (1982).
7
Enzyme Membrane Reactors and

Membrane Fermentors
Enrico Drioli, Gabriele lorio and Gerard0 Catapano
INTRODUCTION
A growing interest in the use of purified enzymes as biocatalysts in labora-

tory scale and industrial applications is being developed as an alternative to more
traditional approaches, generally based on fermentation processes.
The number of papers and patents appearing in this field in recent years
clearly indicate the existence of a new branch of biotechnology, known as “en-
zyme engineering,” dealing with enzyme production, separation, purification
and the development of their applications in enzymatic reactors.
Referring to specific literature ‘r2 for a detailed presentation of enzymes and
their properties, suffice it to say that enzymes are globular proteins which cata-
lyze the complex of biological reactions of a living organism known as “metabo-
lism”. They are usually confined within living cells, often bound to cellular
membranes. Enzymes are organized in colonies and synergistically cooperate;
they catalyze specific reactions contributing to the energy balance of living cells
and at the same time regulate their metabolic state.
Although enzymes play, in a specific reaction, a role similar to that of syn-
thetic catalysts, leading to a substantial reduction of reaction activation energy,
their catalytic action is extremely efficient and selective, well beyond the per-
formances of synthetic catalysts. It is not surprising, therefore, that prospects of
their possible applications has prompted an enormous interest both in universi-
ties and industries leading to a series of related developments, such as:
(a) Enhancement of the techniques to induce microorganisms to pro-

duce selected enzymes;
(b) Improved enzyme purification techniques;
(c) Engineering of techniques to immobilize enzymes or whole cells on

or in solid supports;
401
(d) Development of techniques for enzyme usage in continuous flow

reactors;
(e) Development of techniques toward the solid phase synthesis of

peptides.
Enzymes produced by microorganisms such as fungi or bacteria have been

used for years in batch fermentation plants for the production of pharmaceuti-
cals, beverages, foods, etc. The availability of almost pure enzymes enables one
to carry out specific reactions under mild conditions, to limit side-product for-
mation and to perform the synthesis of chemically active compounds which
would otherwise require extremely long reaction time, with low yields.
The use of enzymes as biocatalysts can therefore be of extreme interest for
individual applications mainly for the advantages it implies in terms of energy
consumption, safety, pollution prevention and materials preservation.
The traditional use of enzymes in batch stirred and similar reactors is how-
ever limited for the following reasons:
(a) High enzyme purification costs;
(b) Low productivity per reactor, per unit time;
(c) Difficult and expensive recovery and reuse of enzymes or cellular

microorganisms;
(d) Product pollution;.
(e) Difficulties in maintaining standard product quality.
Most of the above problems could be overcome using “immobilized en-

zymes”, or in other words “enzymes confined or compartmentalized in a well
defined region of space, which retain their catalytic properties and can be re-
peatedly and continuously used”.3 Several advantages can derive from the use of
immobilized enzymes or microbial cells, such as:
(a) Opportunity to design processes in a more rational way;
(b) Cost cut-off in enzyme consumption;
(c) More compact plants;
(d) Operating costs cut-off;
(e) High productivity per unit time per equipment, with small amounts
of side-products.
A lot of research work has therefore been performed in order to optimize

immobilization techniques and procedures in view of the development of enzyme
reactor engineering.
Traditional immobilization techniques include:
(a) Confinement in a gel;
(b) Encapsulation in a membrane shell;
(c) Adsorption on a surface;

Enzyme Membrane Reactors and Membrane Fermentors 403
(d) Covalent binding to an insoluble support;

(e) Copolymerization with a proteic carrier.
In Japan, a leading country in the development of enzymatic engineering,

more than seventy enzymes have already been immobilized, but only ten of
them are used in most existing applications. This limitation, compared to the
great potential of immobilized enzyme systems, requires that research efforts be
directed towards the realization of more efficient immobilization techniques
characterized by low carrier costs, and the development of practical equipmenL4
Membrane science can strongly contribute, in this sense, to further develop-
ments of immobilized enzyme reactors for industrial processes. This well as-
sessed technology seems, in fact, very promising for the design of compact and
flexible apparatus and particularly useful when a separation step is to be coupled
to the chemical reaction.’ A number of applications of membrane processes to
biotechnology are already in operation.6
Insoluble carrier-fixed enzymes can be continuously used in stirred vessels
or tubular reactors without catalyst loss, and can then be recovered by means of
sedimentors, magnetic fields or similar apparatus.
On the other hand, in nature a continuous uptake of substrate and release of
product without loss of biocatalysts is not achieved by carrier fixation but by
means of cellular membranes. Efficient immobilized enzyme reactor systems for
technical applications can therefore be established replacing cellular membranes
by ultrafiltration or reverse osmosis synthetic membranes, and the activated
transport through the cellular wall by a forced flow across the membrane.’
Ultrafiltration membranes can be used to retain enzymes in the reaction
vessel due to size difference between the usual high molecular weight enzymes
and most of the low molecular weight substrates and products; a homogeneous
continuous catalysis can thus be achieved. Some membrane reactor configura-
tion do not allow a homogeneous distribution of catalyst in the vessel so that
the reactor acts in a more heterogeneous way. Depending on enzyme kinetics,
different reactor configurations exist which are able to optimize system yield;
systems where product concentrations are kept low and a continuous product
removal occur are particularly interesting when dealing with product-inhibited
enzymes, for instance see References 8-26.
Enzymatic systems in which membranes are simply used as separation media
and not as catalyst carriers are traditionally called “enzyme membrane reactors”
(EMR). Concentration polarization phenomena severely affect the performance
of such reactors so that it is necessary to control the polarization layer onto
membrane pressurized side by means of reactor fluid dynamics or design tricks.
Fluid dynamic conditions in some of these reactors make them especially suit-
able for enzymatic systems for which a homogeneous catalyst distribution is
particularly important, such as cofactor-requiring mono- and multi-enzyme
systems.
Concentration polarization phenomena, which are the main drawbacks of
the aforesaid enzyme membrane reactors, can nevertheless be used to form, ei-
ther in dynamic or in static conditions, a gel layer of enzyme proteins on a mem-
brane.2741 It is even possible to establish more than one enzyme layer and no
coupling agent is needed to carry out the immobilization. Due to high protein
concentration on the membrane surface, enzyme stability can be improved over

that in systems using enzymes homogeneously distributed in the reacting solu-
tion. lo6 Reduced catalytic efficiency due to mass transport limitations and the
possibility of preferential pathways in the enzyme gel layer can be typical sys-
tem disadvantages.
Asymmetric hollow fiber membranes can also be used as selective supports
for enzymes. A biocatalyst suspension can in fact be forced through the un-
skinned surface of asymmetric membranes so that biocatalysts, either enzymes
or whole cells, although still suspended, are effectively immobilized within the
macroporous spongy part of the membranes. 42-53 The enzymatic activity can
thus be spread over a large surface, although substrates and products can only
diffuse to and from the biocatalyst.
In certain applications, the size of the biocatalyst is not suitable for entrap-
ment into the supporting sponge layer of asymmetric membranes. In these cases,
the membrane acts mainly as a selective barrier; it defines a reactive zone in the
vessel, usually the shell, accessible to substrates and products mainly by diffusive
mass transfer preventing the catalyst from pollution or inhibition possibly caused
by other species in soIution.54-6g
Enzymes can still be absorbed within symmetric macroporous membranes
in order to establish high catalyst concentrations, cross-linked to prevent them
from elution, or simply covalently or ionically bound either to symmetric or
asymmetric membranes. In spite of short residence times, high conversions can
be achieved in most kinds of enzyme membrane reactors.70q3
Last but not least, viable cells are grown and used in membrane fermentors
for a wide range of applications.“-g8
Our analysis of enzyme membrane reactors will range from systems pro-
posed at bench scale to equipment already in operation at the industrial level,
and from systems where enzymes are bound to membranes to apparatus where
enzymes are simply confined in well-defined regions of the reaction vessel. In the
following sections, we will refer to five main reactor configurations:
(a) Enzyme membrane reactors, where enzymes are continuously

flushed along membranes;
(b) Dynamic enzyme gel layer reactors, where enzymes are immobi-
lized in a proteic gel layer, dynamically or statically formed;
(c) Membrane segregated enzyme reactors;
(d) Membrane bound enzymes in continuous-flow systems;
(e) Membrane fermentors.
LIST OF SYMBOLS
The following symbols are used throughout this chapter.
Membrane surface area, L*
Inner radius of the fiber, L

b (b-a) is the thickness of the dense membrane region
of the fiber wall, L
C Concentration factor
Solid concentration in the gel, M Lv3

%
0 Diffusion coefficient, L2 T-l
D’ Dilution rate, T-l
Oe Enzyme diffusion coefficient, L2 7-l
DL Diffusion coefficient in the bulk liquid phase, L2 T-l
0s Diffusion coefficient in the catalyst annulus, L2 T-l
d Outer radius of the fiber, L
E Enzyme concentration, M Ls3
Ediff Diffusive efficiency
Ekin Kinetic efficiency
EO Initial enzyme concentration, M Lv3
ES Enzyme concentration in the bulk liquid phase, M Lb3
EW Enzyme concentration at membrane surface, M Le3
F Feed flow rate, L3 T-l
J Volumetric flux, L3 Lm2 T-l
K Enzyme kinetic constant, TV1
Kd Enzyme activity decay constant, T-l
K'M Enzyme Michaelis constant, M Ls3
KS Overall mass transfer coefficient, L T-l
1 Reactor length, L
LC Critical reactor length, L
Membrane hydraulic permeability, L2 T M-l

LP
M = KIM/Sf Dimensionless Michaelis constant
M'= K'M/(rR(O) Sf) Dimensionless Michaelis constant for the segregated
region
N Overall enzyme amount in the reactor, M
Nf Number of capillary membranes in a bundle
P Product concentration, M Ls3
Pe = v 6 /D Peclet number
Pe'= vmax a /DI Peclet number
Pf Feed product concentration, M Ls3
PO
Initial product concentration, M LW3
pt- Reactor productivity, or product moles (grams) per
unit time, membrane area (reactor volume) and enzyme
concentration, T-I
Q Flow rate, L3 T-l
Qfl Fiber-to-shell permeate flow rate, L3 T-l
Qf2 Shell-to-fiber permeate flow rate, L3 T-l
R Reaction rate, M T-l Ls3
r Radial coordinate, L
r' = r/a Dimensionless radial coordinate
r"= (r-a)/(d-a) Dimensionless position in the annular catalyst
re Equivalent radius of the shear-free surface, L
S Substrate concentration, M Lm3
Sf Feed substrate concentration, M Lm3
SL = S/Sf Dimensionless substrate concentration in the bulk
bulk liquid phase
SO Substrate initial concentration, M Ls3
SS = S/Q Dimensionless substrate concentration in the catalyst
annulus
SW Substrate concentration at membrane wall, M Ls3
s Laplace coordinate
Tl Downstream time constant (time lag), T
T2 Downstream time constant (segregated region), T
TSS Total suspended solids, M Le3
t Time, T
u(t) Unit step function
V Reactor volume, L3
Vmax -KE Enzyme reaction rate at saturating substrate
concentration, M Le3 T-l
Initial value of V,,,

Enzyme Membrane Reactorsand Membrane Fermentors 407
S
"inax Enzyme reaction rate at saturating substrate
concentration, M LA2 T-l
vss Volatile suspended solids, M Ls3
V Linear velocity, L T-T
v = V/Vmax Dimensionless linear velocity
Vmax Maximum linear velocity, L T-*
x = (Sf_S)/Sf Conversion degree
iT Anaerobic reactor VSS or TSS, M Le3
X' Cell concentration, M Lp3
X'l Cell concentration entering the membrane module,
M L-3
X'P Cell concentration in the recycle stream, M Lm3
X Space coordinate, L
xe Enzyme penetration depth, L
Y Organism yield coefficient, i.e. mass of cells
formed/mass of nutrients
2 Axial distance from the inlet of the reactor to
achieve a given conversion, L
Z'= DL x/(2 vmax a2) Dimensionless axial coordinate
Zdiff Dimensionless reactor length characteristic of
operation under diffusion control
Zkin Dimensionless reactor length characteristic of
operation under kinetic control
L Dimensionless transversal coordinate
2’ = x/(a Pe') Dimensionless axial coordinate.
Subscripts
a, b Refer to region A, B, respectively
1, 2, 3, 4 Refer to region 1, 2, 3, 4, respectively

408 Handbookoflndustrial MembraneTechnology
GREEK SYMBOLS
(I = 6 A/V
(I' Recycle ratio
S
Qo = Vmax /(So v) Dimensionless zeroth order kinetic parameter
S
o1 = Vma,/(K'M v) Dimensionless first order kinetic parameter
B - J /(D/6)
8* = Sf /K’M Reciprocal dimensionless Michaelis constant
7 = s /Sf Dimensionless substrate concentration
7‘ Membrane partition coefficient
7* = K’MDs/[v,,, (d-a)2] Thiele modulus
Dimensionless product concentration

YP = P/P,
Laplace transform of dimensionless product concentration
YP
7s = S/So Dimensionless substrate concentration
6 Gel layer thickness, L
rl Effectiveness factor
9 = t v2 /D Dimensionless time
e* = (8 - '1) Dimensionless time
8' = K'M /Sf Dimensionless Michaelis constant
x Length of the characteristic dimension of the
segregated gel region, L
V a2
1' _( max )W Thiele modulus
KIM 0
Specific cell growth rate, T-l
Solution viscosity, M T-* L-l
Dimensionless space coordinate
Dimensionless space coordinate
Pressure, M L-* Tm2
Pressure of the bulk phase within the capillary
membrane, M L-l Te2
Dbi Pressure of the bulk phase at the inlet of the
capillary membrane, M L-l TW2

Enzyme Membrane Reactorsand Membrane Fermentors 499
Oncotic pressure, M L-l T-*
Shell side pressure, M L-l T‘*
Dimensionless reaction rate,
P = Q1 7s for first order kinetics
P = Qo for zero order kinetics
7 = V/Q Reactor time constant, T
21 = T1 v* /D Downstream dimensionless time constant (time lag)
72 = T2 v* /D Downstream dimensionless time constant (segregated
region)
K E 6*
0 =(- )l/* Thiele modulus
D Sf
K E A2
V'=( )w Heterogeneous gel model Thiele modulus
D 72(O) Sf
6 = 1'2 8'
x = S /-f*(O) Sf Dimensionless substrate concentration within the gel
layer
2
-22 II*
Nz) - e e du Dawson's integral
0 I
o =(d-a) DL/(a Ds) Ratio of diffusion rate in the bulk phase to that in the
porous catalyst
ENZYMEMEMBRANE REACTORS (EMR)
During the last ten years enzyme technology has moved mainly towards
the development of new immobilization techniques and the improvement of
those already existing. In turn, the attention of applied research has been fo-
cused on the engineering of systems based on immobilized biocatalysts. En-
zymes involved in this development were enzymes catalyzing simple reac-
tions that normally require no cofactors. A number of drawbacks affected
the use of immobilized enzymatic preparations. An often dramatic reduc-
tion of initial enzyme activity due to the binding process, and the existence
of diffusional resistances limits this approach with low activity enzymes,
with macromolecular substrates and in general with enzymes whose cata-
lytic behavior is strongly dependent on a good mixing of catalyst and sub-

strate. Biotechnological processes based on immobilized biocatalysts have
not been successful with coenzyme-dependent reactions. Moreover, removal
of deactivated enzymes or cells from the solid support often raises serious
problems and sometimes is not convenient.
Instead of being immobilized, enzymes or cells can be confined in the
vessel of dialytic or ultrafiltration cells, in a configuration that is commonly
called “enzyme membrane reactor” (EMR). An EMR is a reactor system
where ultrafiltration or dialysis membranes with a suitable molecular weight
cut-off are used in order to keep larger components in the reactor vessel, i.e.
enzymes and/or macromolecular substrates, while low-molecular-weight
molecules, e.g., products and/or inhibitors, are allowed to pass freely through
the membrane, thus leaving the reactor as permeate. In this set-up, typical
advantages of immobilized preparations, together with easy recovery of de-
activated enzymes and replacement with fresh catalysts are achieved; inhib-
itors are moreover continuously removed from the reaction vessel. The di-
rect and deep contact between substrates and biocatalysts limits diffusional
resistances, while no activity losses due to fixation to the carrier occur, thus
maximizing the activity of the biocatalyst.
Depending on the flow dynamics of the reaction vessel, it is possible
to distinguish between reactors equipped with flat UF membranes or hol-
low fibers.
So far, both dead-end and CST UF cells with flat membranes have been
proposed as enzymatic reactors (Figure 7.1).
Performance of dead-end units is largely affected by the flow dynamics
of the system; in fact, mixing of substrates and catalysts is not fully accom-
plished, and concentration polarization phenomena strongly limit reactor
performance. On the one hand, the existence of a polarization layer above
the membrane surface significantly reduces permeate flow rates; on the other
hand, enzymatic conversion is achieved mostly inside a highly concentrated
polarization layer giving lower conversions as compared to those obtained
with the same amount of enzyme uniformly distributed within the reactor.
@-
N2
Membrane-l
Figure 7.1: Experimental set-up of an enzyme membrane reactor.” P: peristal-
tic pump; LC: level controller.
CST reactors have been more widely adopted, partly due to the possibility
of concentration polarization control, and partly due to the easy modeling of
enzyme kinetic behavior. In the literature, comprehensive mathematical de-
scriptions of the kinetic behavior of enzymes located in CSTR UF units are re-
ported8t9r’e.as far as low-molecular-weight substrates are concerned.
Assuming complete mixing within the reactor so that enzyme and substrate
concentration in the reactor vessel are uniform, and the latter is equal to its value
in the permeate, substrate mass balance in molar form generally looks like:
(Sf-S)/T = R(S,P) (I)
Product steady state mass balance similarly appears as:
(P - Pf)/r = R(S,P) (2)
Assuming that substrate conversion obeys the simple Michaelis-Menten

model, substrate steady state mass balance reduces to:
XSf + [X/(1 - X)1 KIM = KEV/Q (3)
R being expressed as K E S/(K’M + S).

The equation in this form is a useful tool to estimate parameters of reac-
tion kinetics. Instead of performing nonlinear parameter estimation procedures,
the functional dependence of XSf on X/(1 - X) can be plotted. The plot should
look like a straight line whose slope is (-KIM) and whose intersection with the
XSf axis is given by the point of coordinate (Vmaxr). Hence, it furnishes a rapid
graphic procedure to obtain rough estimates of kinetic parameters.
When more complex kinetics are involved, so that the substrate rate of con-
version is dependent on both substrate and product concentrations, combining
Equations 1 and 2 with the rate equation eventually leads to an implicit Equa-
tion:9
(P - Pf)/r = R(P) (4)
Product concentration in the permeate can then be determined either with

numerical procedures or with graphic techniques. True reactor operating point at
steady state is given by the intersection point of the straight line representing
mass balance and the curve of the reaction term.st10r99As can be seen from Fig-
ure 7.2, for cu-ketoisocaproate conversion to L-leucine, the slope of the convec-
tive term is reciprocally proportional to the mean residence time. If multien-
zymatic systems are used as the biocatalyst, it must be outlined that the reac-
tion term is the course of the reaction rate relative to the substrate conversion
of the key component.”
Enzyme activity is usually not constant with time.
Physiochemical changes in enzyme structure, thermal denaturation and mi-
crobial contamination cause enzyme activity to continuously decrease with
time. When enzymes or cells are compartmentalized in UF cells, biocatalyst
losses can even occur due to the wrong choice of membrane molecular weight
cut-off. It is conventional to measure the enzyme stability in terms of its half-
life, tx, that is the time at which enzyme activity is reduced to half its initial
0.4
X
0
7-g
lG 0.2
0.0 i
1 0.8 CL6 0.4 02 0

Substrate conversion
Figure 7.2: Relative reaction rate of the formate dehydrogenase-leucine dehy-

drogenase (FDH-LEUDH) system as a function of substrate conversion.”
FDH = 20 U/ml; LEUDH = 20 U/ml.
value. Since biotransformations by means of enzymes are continuous processes,

as long as the reactor working life is longer than the native enzyme half-life, en-
zyme activity decay with time must be taken into account in order to correctly
assess reactor performance, or a strategy in reactor operation. A transient sub-
strate mass balance on the CST reactor leads to the Equation:s
V[dS(t)/dtl = Q[Sf - S(t)1 - V,,,V (5)
Let us assume that substrate inlet concentration is much higher than the
enzyme’s apparent Michaelis constant, K’M, so that enzyme kinetics is ex-
pressed according to zero order kinetics, i.e., R = KE = V,,,.
Enzyme activity decay can be expressed in terms of the Arrhenius Equa-
tion :
E = E,exp(-Kdt) (6)
where Kd is the deactivation constant. Equation 5 then takes the form
[Sf - S(t)1 /r - dS(t)/dt = V,,,, expf - Kd t) (7)
Integration of the differential Equation 7 leads to estimate the outlet sub-

strate concentration:*
S(t) = v,,, [exp( - t/r) - exP( - Kd t)l /(l/r - Kd) + Sf (8)

Figure 7.3 shows good agreement of experimental data and theoretical pre-
dictions for benzaldehyde conversion by Rhodotorula Mucilaginosa.8
For 057, exp( - t/r) contribution can be neglected so that enzyme half life
time, tr/, can be expressed as:
tr/, = InjV,,o/I(1/7 - Kd) (Sf - So)/2jl/Kd (9)
1.6
0.6
01 I I I I I 1
0 10 20 30
t, d
Figure 7.3: Experimental and (solid lines) theoretical plots of substrate concen-
tration (S) as a function of time, for various retention times: (A) 0.417 days,
P) 0.520 days, (0) 0.695 days8
Estimate of the deactivation constant can be graphically carried out from

Equation 7 plotting In [[Sf - S(t)1 /T - dS(t)/dtl vs t (Figure 7.4). A straight
line is thus obtained whose slope is given by ( - Kd) and the line intersects the
vertical axis at the point of coordinate (In V,,,,).
When the reacting solution is fed to the system, low-molecular-weight prod-
ucts leave the reactor permeating the membrane, whereas enzymes are partially
or totally rejected. Then, enzymes tend to accumulate in a thin layer immedi-
ately upstream from the membrane causing polarization phenomena to occur.
The extent to which concentration polarization affects reactor performance
depends on the balance of rejected solutes, e.g., enzymes or cells, accumula-
tion due to membrane rejection and back diffusion to the bulk phase, and even-
tually on the flow dynamics of the reacting vessel.
For macromolecules (like most biocatalysts), if membrane properties are
carefully chosen, membrane rejection is usually very good, while biocatalyst
0.8.
0.6
0 10 20
14
Figure 7.4: Values of Vmaxo and Kd for various retention times: (A) 0.695
days; (B) 0.520 days; (C) 0.417 days. Straight lines are approximations of exper-
imental data Ito a straight line calculated from Equation 7.8
back diffusion towards the bulk phase is extremely slow. The effect of concen-
tration polarization on such reactor performances can then be significant. Fig-
ure 7.5 shows a schematic of the situation. Applying the thin film theory to a
region immediately upstream from the membrane results in the following
steady state mass balance equation:‘w~lO1
J E = - D,dE/dx (10)
under the B.C. x + m E = Es.

Under the assumption of totally rejected enzyme macromolecules, in-
tegration of Equation 10 allows one to express the ratio of enzyme concentra-
tion in the bulk solution in the presence of concentration polarization, to its
1 J*A
Figure 7.5: Schematic diagram of enzyme membrane reactor.’
value in the absence of concentration polarization phenomena as:s
ES/E, = I/[(1 -a) + (a/P) [expW)- 111 (11)

Enzyme concentration at the membrane-solution interface can thus be re-
lated to enzyme concentration in the bulk phase by the following equation:
EVV = Es expU3 (12)
where the dimensionless parameter fl is a measure of convective mass transfer,

J, relative to the overall mass transfer, D/6, more generally Ks.“* Figure 7.6
shows how dramatic the change of enzyme bulk concentration can be, with
changes in permeate flow rate, and hence applied pressure. Doubling flow rate
from 4 to 8 ml/min results in an B-fold decrease of fractional bulk concentration
of soluble enzyme, from 97% to 12% for the cellobiose/cellobiase system.
Changes in enzyme bulk concentration may induce a large reduction in re-
action rate within the membrane reactor. Figures 7.7 and 7.8 give evidence of
the existence of enzyme deactivation phenomena which can be misleading in
the evaluation of reactor performances, and/or of the thickness of the bound-
ary layer close to the membrane surface in accordance with Equation 12.
0.8
4
10.6
IA?
0.4
0.2
\
0.0 1111111111
0 2 4 6 8 10
Flow Rate, ml/min
Figure 7.6: Dimensionless bulk concentration (Es/E,) versus flow rate for a
membrane reactor with a PM 10 membrane.s V = 300 ml. A = 39.2 cm’, D =
6.8 x lob7 cm*/s. The parameter is gel thickness times 1,000 cm.
Flow Rate
z
21.6 - 2.5 - .3 ml/min
J
E, 0
r:
E
s
c 0.8 B
*-----------
0”
%
%
! AkA
s 0.0 e L 1
0 8 16 24 32
Time, hr
Figure 7.7: Concentration of glucose in permeate versus reaction time:(o) at

flow rate of 2.5 ml/min, solid line is expected level for ideal operation (1.37
mg/ml); (A) at flow rate of 7.3 ml/min; dotted line is expected level for ideal
operation (0.77 mg/ml).s
A A &A A
16 24 32
Time,hr
Figure 7.8: Concentration of glucose in permeate versus reaction time: (0) at
flow rate of IO.3 ml/min; solid line is expected level for ideal operation (0.8
mg/ml); (a) at flow rate of 8.05 ml/min; dotted line is expected level for ideal
operation (1.21 mg/ml).’
Under given stirring conditions, therefore, a critical value of flow rate

and hence of applied pressure does exist. At flow rates lower than the
critical value, the reactor always attains steady operation conditions; and
correspondingly outlet product concentration is constant. Beyond the
critical value, concentration polarization phenomena promote the localiza-
tion of a large fraction of enzymes near the membrane surface, seldom lead-
ing to an enzymatic gel formation.’ Correspondingly, an accelerated deac-
tivation of enzyme activity is superimposed on reactor performance thus
hindering the attainment of steady state conditions. Experimental evidence
suggests that the contribution of polarized enzymes to overall conversion
can be negligible, due to the consistency of product concentration in the
bulk phase of the reactor and in the permeate. As suggested by Tsao et al,9
a continuous conversion decrease, as concentration polarization occurs, can
be attributed to the localization of large amounts of enzymes in a region
immediately upstream from the membrane; in a stirred UF unit proteic
coiled macromolecules in the polarization layer are subjected to a high shear
field due to stirring, which can result in protein unfolding and consequent
denaturation. Overall enzyme deactivation is moreover accelerated by the
exchange of deactivated enzymes in the polarization layer with fresh en-
zymes in the bulk solution via a coupled diffusion-convection mechanism.
Irreversible enzyme deactivation is confirmed by the discrepancy between
the expected values of conversion and experimental data as applied pressure
is slowed down from hypercritical to subcritical values (Figure 7.8). Bound-
ary layer thickness under such conditions cannot be easily estimated from
the conversion at various permeate flow rates. Enzyme deactivation, being
superimposed on the enzyme concentration decrease in the bulk phase, can

be misleading. It is to be noted that deactivation phenomena are less promi-
nent for enzymes stable to shear stresses.
When macromolecular substrates are involved in the transformation un-
der study, concentration polarization phenomena affect the EMR perform-
ance more severely. Diffusion limitations of macromolecular substrates
hamper the use of immobilized enzymes in the hydrolysis of high-molecu-
lar-weight substrates. By selecting membranes with an appropriate molecu-
lar weight cut-off, both enzyme and substrate are retained in an EMR in
touch with each other, and hydrolysis products and/or inhibitors are contin-
uously removed from the system. Soluble enzymes can then act directly on
substrate macromolecules without diffusion limitations and steric hindrance
imposed by enzyme fixation to a solid support. The stirring features of CST
EMRs moreover assures that substrates and/or inhibitors within the reactor
vessel are maintained at the lowest possible concentration level. Such reac-
tor configuration is then extremely useful when substrate inhibited reaction
patterns are involved, or when inhibiting species are assumed to exist in the
feed stream.
Diafiltration,” semicontinuous’ and continuous operational modes13-15
have been proposed and applied for saccharification of cellulose11~*2~14~15
and protein hydro1ysis.a As far as the first two operational modes are con-
cerned, Figure 7.9 shows typical curves of product concentration as a func-
tion of time for the hydrolysis of cellulose according to a diafiltration oper-
ational strategy, at different space velocities. A peak is always observed in
the first period of operation, more marked for lower residence times, followed
by a slow decay. Initial sudden increase in product concentration in the per-
meate occurs earlier at lower residence times, i.e., higher flow rate. Such find-
ings confirm that enzymatic hydrolysis of cellulose is strongly product inhibited
and that suitable flow dynamic conditions greatly improve reactor performances,
as shown in Figure 7.10.
Operating the reactor in a semi-continuous condition and adding sub-
strate so as to keep its bulk concentration constant, leads to meaningful
changes in reactor performance as compared to the diafiltration mode. Un-
der either operational mode permeate flow rate continuously decreases
with time.
When EMRs are operated continuously, feeding a slurry of substrate macro-
molecules to the reactor, concentration polarization phenomena play a domi-
nant role. The presence within the reaction vessel of contaminants or intermedi-
ate products, e.g., dextrins in starch hydrolysis,r3 which are not fully hydrolyzed
by the enzymatic system under study can lead to membrane foulingI or to the
formation of a gel layer at the membrane surface. Under such conditions, the
filtration rate continuously decreases with time, and it may happen that, after a
sudden increase in the early stages, substrate conversion does not attain steady
state conditions, Figure 7.11. The addition of enzymes capable of hydrolyzing
such foulants to low-molecular-weight compounds usually improves reactor per-
formance, eventually approaching steady state conditions in terms of both per-
meate flow rate and substrate conversion.16j’7 In addition, antifouling proce-
dures including suitable feed pretreatment or procedures to reduce concentra-
L l
4 8 12 16 20 24 28 32 36 40 44 4a
TimeIhr)
Figure 7.9: Concentration of soluble sugars (cellobiose and glucose) in permeate

from an enzyme membrane reactor as a function of time for various space ve-
locities (S).”
Conversion(%)
60 I
4 8 12 16 20 24
Figure 7.10: Conversion of cellulose to soluble sugars as a function of time for

various space velocities (S).”
tion polarization can be used. lo1 Equation 12 demonstrates the dependence

of enzyme concentration at the upstream membrane surface on the flow dy_
namics of the reaction vessel. Due to the monotonicity of the exponential func-
tion, at given applied pressure, i.e., J, operational conditions improving mass
transfer in the bulk phase, i.e., KS or D/6, diminish concentration polariza-
tion.“’ Figure 7.12” shows how high stirring speeds usually improve filtering
performance of a CST flat membrane reactor, albeit at the expense of partial
enzyme deactivation, as discussed earlier.
g 24
s
g! 18
2
s
‘3
6 12
E
6
220
e
!j 140
3
E,
60 t 1 I
0 25 50 75
0
Figure 7.11: Effect of adding crystalline sweet potato&amylase (87 c(gprotein/ml
reaction mixture) on the filtration rate (ml/h) and on the reducing sugar content
(measured as maltose) in the filtrate at 50%. 15 psi, in a stirred cell system. En-
zyme added (@I; no enzyme added (o).13
0.10. r I I I
2 UM-IO MEMBRANE
M- 50 CELL
0.65% PROTEIN lItI RPM)
TRANSMEMBRANE PRESSURE (psi)
Figure 7.12: Flux-pressure relationship for bovine-serum albumin solutions in a

stirred batch cell at various stirring rates and protein concentrations.“’
Reactor configurations other than dead end or CSTR can offer improve-
ments. In Figure 7.13, the performance of a thin channel flat membrane reac-
tor for the hydrolysis of sweet potato by means of a-amylase is reported in
terms of filtration rate and product concentration vs time plots. A comparison
to Figure 7.11 shows how great the improvements due only to the change of
configuration can be. At an appreciably high flow rate and product concentra-
tion level, a steady state is also attained.
The use of hollow fiber ultrafiltration modules strongly improves both re-
actor performance and economics. Capillary membranes are characterized by
a favorable surface to volume ratio; advantages from this feature are not only
related to lower overall plant size, but also to the increase of the surface to price
ratio. Flushing enzyme/substrate solution from a reservoir to an HF module in
a semiclosed loop at a high linear velocity reduces the extent of concentration
polarization avoiding the formation of a secondary enzymatic gel membrane, in
spite of product flux through the membrane. 101t102Provided that the volume of
the UF module is small relative to the total volume, and that recirculation flow
rate is much larger than permeate flux, system kinetic behavior can be modeled
in terms of a CST reactor. Assuming biocatalyst kinetics are described by the
simple Michaelis Menten model, the substrate mass balance can be expressed ac-
cording to Equation 3. ‘s Reactor productivity as a function of time can be re-
lated to flux and enzyme concentration according to the following equation:
Pr = x(t) SfJ(t)/E V (13)
300
it! 32.0
160 Se-l% w/v
sii V =550 ml
y 200
E =(1066gm -
t // nn a E =0.31& gm _____
s
In
.=
; 100
.-
t;
r/( A, _*.____-““-3
f!
a 0 .I
200 400
Time , min
Figure 7.13: Effect of adding crystalline sweet potato /3-amylase (87 pg pro-
tein/ml reaction mixture) on the filtration rate (ml/h) and on the reducing sugar
content (measured as maltose) in the filtrate at SO%, 15 psi, in a stirred cell
system. Enzyme added (0); no enzyme added (o).r3
Figure 7.14 shows how reactor productivity (g product/g enzyme) is af-

fected by flux and enzyme concentration, for soy protein hydrolysis performed
by Pronase enzyme. Maximum productivity can be obtained by operating the
reactor at the highest flux and the lowest enzyme concentration.
Figures 7.15-7.18 show how reactor capacity (i.e., maximum productivity,
g product/g enzyme/min) is affected by operational parameters. Enzyme and
substrate concentration appear to affect reactor capacity to a large extent
(Figures 7.15 and 7.16) especially at lower enzyme concentrations. Effects of
permeate flux and reactor volume on conversion are related to each other by
means of substrate residence time in the reaction vessel. Increasing residence
time, that is increasing reactor volume or lowering fluxes, results in higher con-
versions at the expense of lower capacities (Figures 7.17 and 7.18).
Comparisons of EMRs operated in the diafiltration mode” and in the con-
tinuous mode” to batch reactor performances are reported in Figure 7.10 and
7.19. Figure 7.10 shows how conversion rate is improved by continuous removal
of inhibiting products. Figure 7.19 (reactor continuously operated) shows that
productivity levels are higher the longer the reactor is operated before shutting
down for cleaning, recharging or backflushing. In the continuous process, en-
zymes are charged at the beginning of the process, and the system is kept work-
ing until enzyme activity drops down to a given limit established on the base
of enzyme half life and plant economics. In batch reactors, the enzymes must
be replaced at the end of each cycle. Hence the longer the reactor is operated,
the greater is the productivity and the larger the difference between continuous
and batch system performance. Moreover, batch processes require added ex-
penses for enzyme inactivation and product purification, as well as more labor.
E =Q066gm _
E =Ct31C gm _____
200 400
Time, min
Figure 7.14: Effect of flux and enzyme concentration on UF reactor produc-
tivity. E = weight of enzyme added to reaction vessel (flux at E = 0.314 g: (0) =
9 ml/min; (A) = 16 ml/min).”
Conversion
+ 1% w/v
J = 90 ml/min
0
0 0.4 0.8 l.2 1.6
Enzyme Concentration, mg/ml
Figure 7.15: Effect of enzyme concentration on conversion and capacity of a
UF reactor. Pronase-Promine D at pH 8.0, 5O’C.”
E = 0.12 mg/ml
J = 90 ml/min
V = 550 ml
E
08 100
‘i, 1.6 z?.
.
g
._
u)
L
.-
2
$j- 0.8
0
0 1 2 3
Substrate Cont. % w/v
Figure 7.16: Effect of substrate concentration on conversion and capacity of a

UF reactor.lg
Handbook of Industrial Membrane Technology
2.4
0 10 20 30
Flux (ml/min)
Figure 7.17: Capacity and conversion of a UF reactor as a function of flow rate
through the system (flux).”
Conversion
0 1000 2mo 3ooo
Figure 7.18: Effect of reactor volume on UF reactor capacity and conversion.‘g
The real potential of membrane reactors becomes evident with coenzyme

dependent enzymatic systems.‘0~20~21Coenzymes, like NAD’ or NADP’, usually
have a long term effect on enzyme activity only if they can move from one en-
zyme, able to oxidize them, to another, able to reduce them, in loop kinetics.
Continuous homogenous catalysis is then a prerequisite to achieving high reac-
tion yields. Enzyme membrane reactors offer a suitable reaction environment
provided that coenzymes are retained in the reaction system. Such compounds
are in fact quite expensive, which limits the use of coenzyme dependent en-
zymes. Reverse osmosis (RO) membranes could be helpful in retaining native
‘-“““““““““‘-‘T’
0 2 6 8 10
Volume Replacements, J f/v
Figure 7.19: Comparison of productivity of a continuous UF reactor and a
batch reactor for the Promine D-Pronase system at pH 8.0. Volume replacement
for the continuous UF reactor calculated as J t/V.19
coenzymes in a given vessel on the basis of their size. Unfortunately, permea-

bility of such membranes towards water is poor as compared to UF membranes,
thus limiting overall fluxes. Moreover, the presence of product in the permeate
implies coenzyme loss due to the usually small differences of their molecular
weight. In the early 197Os, some procedures were proposed in order to in-
crease coenzyme molecular weight by covalently binding the coenzyme to a high-
molecular-weight soluble polymer such as polyethyleneglycol ordextran.‘0~‘03~104
Choosing the membrane molecular weight cut-off so as to reject the soluble
polymers results in retaining the coenzymes within the reaction vessel. If a suit-
able pair of enzymes is charged into the reactor together with the macromo-
lecularized coenzymes, the latter are continuously regenerated and long term
operation is assured. Laboratory and plant scale experiments”t*’ show the im-
portance of a careful choice of the enzymatic system, as well as the ratio of oxi-
dizing to reducing enzymes. Differences in enzyme cofactor requirements, half
life, and optimum pH and temperature can in fact render one enzymatic system
rate limiting thus reducing reactor performance.
Patents already exist such as for amino acid production from a-ketoacids,**
which confirm the potential of scale-up for EMR plants. For this particular proc-
ess, resolution, derivatization and racemization steps can be avoided, thus reduc-
ing overall costs.
EMRs have a number of advantages over other immobilized preparations. In
summary:
(a) Homogeneous catalysis

(b) No diffusion limitations
(c) No activity lossesdue to carrier fixation

(d) Absence of enzyme fixation costs
(e) High activity per unit volume
(f) Sterilizability of the plant
(g) Constant productivity assured by enzyme dosage
(h) Absence of contaminants
(i) Interchangeability of substrate/enzyme systems
(j) Use of multienzyme systems.
Efficient circulation, however, is required to prevent concentration polar-

ization, which becomes quite a problem as enzymes are added to keep con-
version constant. A large membrane area and sterile filtration are also needed
to assure high yields over long periods of time. If the enzyme solution is to be
flushed at high linear velocity near the membrane surface, shear forces may de-
activate enzyme molecules.
Thus far, EMRs have been successfully used with macromolecular sub-
strates, as for the saccharification of cellulose11-13~15~26
and protein hydrolysis,”
or with low molecular weight substrates, as for cellobiose hydrolysis’ and L-
malic acid production from fumaric acid.*’ Bench and large scale plants are al-
ready in operation for the preparation of N-acetyl-D,L derivatives from L-
amino acids by means of acylase and the production of L-malic acid from fu-
maric acid by means of fumarase.*’ In the case of fumarase, conversions of up
to 86% with resolution rates of up to 85% have been attained.
In conclusion, flexibility of such systems can be considered to be limited
only by the number of active compounds which can be synthesized via an en-
zymatic route.
DYNAMIC ENZYME GEL LAYER REACTORS
Ultrafiltration of protein solutions is a proven unit operation”’ for ob-

taining purified enzymes from cell cultures. Under certain circumstances, such
processes can provide an interesting technique for enzyme immobilization which
takes advantage of concentration polarization phenomena. If an enzyme solution
is flushed under pressure through a UF membrane that completely rejects en-
zyme molecules, these molecules will accumulate on the active membrane sur-
face and possibly deposit there as a thin gel layer characterized by enzymatic
catalytic activity.*‘r** Actual gelation of enzyme proteins, and hence their dy-
namic immobilization, depends strictly on enzyme concentration at the mem-
brane-liquid interface,29 (F’g I ure 7.20). When the maximum enzyme concentra-
tion is lower than the gel concentration value, enzymes are not immobilized. Al-
though they are confined near the membrane surface at fairly high concentra-
tion levels, they are still in soluble form.30
If gel formation occurs, enzymes are effectively immobilized without mean-
ingful changes in their microenvironment.31 Moreover, their settlement is partic-
C “bulk
bulk
membrane gel-layer
Figure 7.20: Gel layer formation ultrafiltering a protein solution through a

semipermeable membrane. x is the distance from the membrane wa1l.31
ularly attractive due to the high enzyme and protein concentrations in the gel,
the latter strongly enhancing enzyme stability.lo6
Enzyme gel layers can be built up under a number of fluid dynamic condi-
tions. Unstirred and stirred batch reactors have been used as well as systems
where the enzymatic solution is kept continuously flowing along semipermeable
membranes until gel formation sets in.
When unstirred batch membrane units are used as reacting vessels,a steady
state mass balance on retained species, i.e., enzymes, leads to the evaluation of
their concentration as a function of the distance from the membrane surface, x.
Taking into account both convective and diffusive mass transfer mechanisms,
the enzyme mass balance equation at steady state is as follows:
Ded2E/dx2 + V dE/dx = 0 (14)
Proper boundary conditions can be derived assuming (a) complete enzyme

rejection and (b) no enzyme loss, that is:
B.C.1 x=0 DedE/dx+vE=O (15)
B.C.2
E(x) A dx = N
Upon integration of Equations 14 and 15, the enzyme concentration pro-

file appears to be:30r32
E(X) = [Nv/(ADe)I exp[-vx/(De)l (16)
Maximum enzyme concentration occurs at the upstream membrane surface,

(for x = 0) and it can be expressed as:
Ew = Nv/(ADe) (17)
Enzyme depth of penetration, defined as the ratio of gel volume (in which
99% of the initial amount of enzyme is contained) to membrane cross-sectional

area, can be estimated to be:30
xe = 4.6 De/V (18)
Enzyme average concentration in the gel is therefore 21% of enzyme con-

centration at the gel-membrane interface.
With the enzyme gel concentration known, Equation 16 can be used as a
design equation in order to estimate the amount of enzyme, N, required to build
up an enzymatic gel layer of thickness x.
Similar equations under different fluid dynamic conditions can be derived
from Michaels’ gel formation theory”’ or from similar theories modeling con-
centration polarization phenomena.10’~102~‘07~108
Unfortunately for most substances, an accurate knowledge of gel concen-
tration is not available, and only order of magnitude estimates can be gained
from the literature.“’ Protein gelation, for instance, is thought to occur at con-
centration levels ranging from 20% to 40% by weight. Criteria need to be devel-
oped in order to experimentally detect enzyme gel build-up.
Ultrafiltration of protein solutions usually results in a time progressive de-
cay of permeate flow rate, which after a period of time attains a steady state
constant value for a given transmembrane pressure. Such a decay can be attri-
buted either to gel formation, or to an increase in osmotic pressure of the ultra-
filtered solution as a result of macromolecule accumulation at the membrane-so-
lution interface which lowers the pressure driving force. If a gel is formed, any
further transmembrane pressure increase does not enhance permeating fluxes
correspondingly. In some cases,3o permeate flux decay associated with gel for-
mation is negligible thus ruling out the possibility of detecting eventual gel for-
mation.
The steady state kinetic behavior of such an enzyme membrane reactor is
only related to the kinetic behavior of enzymes in the gel or in the soluble state.
Unless knowledge of intrinsic enzyme kinetics both in the gel and in the soluble
form is available, a steady state analysis is not useful for determining whether
gel formation has occurred.
An effective distinction between enzymes in gel form and enzymes in the
soluble state can be achieved by the analysis of the transient behavior of the en-
zymatic reaction.30t33
Figures 7.21 and 7.22 show typical results for the two cases-with en-
zymatic gel layer formation and when soluble enzymes are confined only near
the membrane surface. Comprehensive models for an immobilized enzyme batch
membrane reactor (IEMR) and for a soluble enzyme batch membrane reactor
(SEMR) are proposed in References 33 and 30, respectively, for a flat slab
membrane configuration.
As far as an IEMR is concerned, analytical closed-form solutions are avail-
able for limiting first and zero order kinetic rate equations.33
The unstirred reacting vessel is divided into two sections-upstream and
downstream of the membrane supported enzyme gel layer. Referring to the up
stream region, the mass balance must take into account both diffusion and con-
vection due to the usual low permeating fluxes. Assuming that neither reactant
nor product is rejected by the membrane and that diffusion coefficients of
% 24r
0 40 80
t, min
Figure 7.21: Hydrolysis of 1.5 mM sucrose by cogelled invertase in an immobi-
lized enzyme membrane reactor (IEMR). Comparison between experimental
data in terms of glucose outlet concentration (C,) vs. reaction time (t) and theo-
retical predictions (-1. Enzyme amount, 0.398 x IO-’ g/cm2; reaction tempera-
ture, 30°C; pH 4.65; flow rate 1.13 x IO” ml/s.30
^/
. . .
.. la
01 1
0 200 400 600 800 IO00

t (min )
Figure 7.22: Hydrolysis of 1.0 mM sucrose by soluble invertase in a soluble

enzyme membrane reactor (SEMR). Comparison between experimental data
in terms of glucose outlet concentration (C,) vs. reaction time (t) and theo-
retical predictions (-). E = 1.06 x IO-” mols/ml; Sf = 10 mols/ml, V = 70 ml,
flow rate = 1.6 x 10m2ml/s, T = 3O”C, pH = 4.65.30
either compound are constant and close to each other, mass balance equations
in dimensionless form can be written
-+-=- ays ays (19)

x2 ac ae
(20)
with the following boundary conditions:
e=o Y, = 1 ; Yp = 0 (21)
c-+- y/1 ; y =o
P
B.C.2
Boundary condition B.C.1 states that the height of the region where sub-
strate and product concentrations appreciably differ from their feed values is
negligible compared to cell height. B.C.2 stems from the assumption that en-
zymatic reaction takes place only on the enzyme layer free surface.
The downstream region of the UF test can be modeled in terms of a first
order system in series to a pure time lag. Dimensionless product concentration
at the UF cell outlet is then related to that immediately downstream of the
membrane by the system transfer function.
Yp (s) e
- Tl s
(22)
uP (0,s) (1+ ‘c* s >
Overall system response in time domain can then be evaluated from the
system response in Laplace domain for first and zero order kinetics:
_-
e*
CL1 + 0.5 0.5
Y,(e) = I ( 1 - e r2 ) -- erf J 0.25 O* + @CO*)+
al+ 1 “1 + 1
(23)
*
A
--
(a1 + 0.5) e *2
+
3 1 l
T2 (al+l).( 01 +a1 +-_)

72
2 1
((al + a1 + - 1 0”)
. (1 - e T2
erfc ((al + 0.5) GF ) I “M”)
and
f
Y,(e) p: a0 IO.5 ( ~2 - e*> erfc ’ 0.25 e* + erf ’ 0.25 e +
_- e* (24)
-0.25 e*
12
+Ke - 0.5 T2 e +r(e*) 1 u(e*)
respectively, with
--e*
T2
erf((( 0.25 -+ e”? )
7 e
o(e*) = (0.25 - +”
2
(a1+al+J--)
0.25 ~~ -- e*
72
‘u(e*) = e erf (( 0.25 - + )$ e* ‘I 1
( 0.25 - + )+
and (25)
- 0.25 f3*
2al(-!-- 0.25 )$ e
T2
o(e*) = - 11((( + - 0.25 )e* ) ‘I )
G ( q2+ al +
+
T2 < 4
- 0.25 e*
0.5 e
Y(e*) - 0.25 ) 8* ) ’ )
(‘_
‘52
Figure 7.23 depicts a model fitting experimental data for the enzyme in-
vertase for both kinetic orders3’
An overall steady state mass balance relates kinetic parameters of either rate
equation to the ultimate value of the dimensionless product concentration yield-
ing:
ai
lim Y,(9) = (26)
e+m ( al + 1)
lim Yp(e) = a0
e+m
When the amount of enzyme is insufficient to ensure that the enzyme con-
centration at the membrane surface attains the gel value, Equation 17 indicates
that the enzyme is confined in soluble form at a fairly high concentration in a
region immediately upstream from the membrane. A soluble enzyme membrane
reactor (SEMR) is thus defined. As soon as the substrate stream is fed to the sys-
tem, enzyme remixing occurs throughout the vessel and reaction then starts
throughout the whole cell. As time goes by, enzyme macromolecules are pres-
sure driven towards the membrane surface; thus reaction rate in most of the cell
volume progressively decreases due to an enzyme concentration decrease. Im-
mediately upstream of the membrane, the reaction rate increases due to an en-
zyme concentration increase. Once steady state conditions are attained, reac-
tion takes place only in the region where enzymes are confined and whose thick-
ness is determined by Equation 18.
-zero order
..* . . *
.
0, I 1
0 20 40 60 80
t , min
Figure 7.23: Hydrolysis of sucrose (3 x IO” mols/ml) by immobilized invertase

operating in an IEMR. Comparison between experimental data (expressed in
terms of glucose, C,) and theoretical curves (solid lines) calculated according to
Equations 23 and 24.32
Simple mathematical modeling of this situation is reported in Reference 30

referring to the equivalent lumped parameters system in Figure 7.24. Also in
this case the reacting volume is modeled as two separate regions in series to each
other. Region A, where a time dependent enzyme depletion occurs, virtually co-
incides with the cell as a whole except for the negligible liquid volume where
enzymes accumulate, namely region B. Each region is modeled in terms of a sin-
gle well mixed tank. As far as region A is concerned, for a particular UF test cell
this assumption can be verified simply by a tracer step response test performed
on the system itself.
Assuming a uniform enzyme distribution within the cell volume at system
start-up, the functional dependence of enzyme concentration in region A on the
reaction time is obtained upon integration of the mass balance equation:
E,= 7a dE,/dt (27)
subject to the initial condition
I.C. t = 0 Ea = N/V,
to give
Es(t) = (N/Va)exp(- t/ra) (28)
Regarding region B as a lumped parameter system of volume Vb (with

Vb=xe A) fed by the outlet from region A, the average enzyme concentration
Actual System Lumped Parameters

System
Figure 7.24: Actual and lumped parameters system. VA is the dominating vol-
ume portion where enzyme depletion takes place; Vs is the constant volume in
which enzyme accumulation occurs.3o
in region B can be expressed as follows:
Eh(t)=(k/V,) [I+ ; (l-exp(-t/~,)l (29)
Similarly, substrate mass balance in region A looks like

dSa
Sf - Sa - Ra 7a = Ta --dt- (30)
The kinetic equation describing region A, Ra, can be assumed identical to

the rate equation for enzymes in homogeneous solution.
Referring to region B, it has been shown33 that the ratio between substrate
and enzyme depths of penetration, xe, is equal to the ratio between the corre-
sponding diffusivities. In most cases, enzyme depth of penetration, i.e., thickness
of region B, can then be assumed to be one order of magnitude lower than that
for the substrate. Substrate concentration is, therefore, virtually constant over
the whole region B. Hence, substrate mass balance in this region is expressed ac-
cording to the dynamics of a stirred tank reactor of volume Vb, that is:
dSb -
Sa - Sb - Rb Tb = Tb dt (31)
An overall product mass balance at steady state yields the proper rate equa-
tion for region B:
Rb = (Q/vb)pb
Pb being product concentration in the permeate. Once steady state conditions

are attained, all enzyme is virtually confined within region B at a concentration
Eb=N/Vb. Therefore, enzyme kinetics in region B can be experimentally de-
termined through Equation 32. For details on the initial conditions and in-
tegration of the set of Equations 30 and 31 readers are referred to Reference 30.
Ultrafiltration of an enzyme solution through a UF membrane does not al-
ways result in gel layer formation. Unless a gel layer is formed, this immobiliza-
tion technique cannot be used for flow systems lacking effective enzyme immo-
bilization. In any event, soluble enzyme membrane reactors can be useful in
order to perform kinetic analysis at high enzyme concentrations. Once steady
state is attained, the theoretical model permits calculation of reaction rates in
both regions. Once the layer is formed it behaves like a secondary membrane,”
capable of separating compounds of different molecular weight in the mixture
as well as catalyzing a chemical reaction.
Enzyme immobilization via dynamic formation of an enzyme gel layer has
been applied both to flat and tubular membrane reactors, either recirculating
the permeate or the axial stream.
The kinetic behavior of these immobilized systems is to be analyzed taking
into account:
(1) The rate equation for native enzymes and for enzymes in gel form
is not necessarily the same due to microenvironmental and shear
stress effects;
(2) Within the gel layer, substrate mass transfer and reaction occur si-
multaneously giving rise to substrate concentration profiles at lev-
els lower than the feed concentration;
(3) External mass transfer resistances have to be taken into account oc-
casionally, depending on operating conditions and reactor config-
uration.
Experimental kinetic data at different temperatures and at saturating sub-

strate concentrations can be used to evaluate the relative importance of all the
aforesaid phenomena.
Plotting such data in terms of the logarithm of the specific reaction rate vs
the reciprocal of the absolute temperature (i.e., an Arrhenius plot) is helpful in
assessingwhich step is rate controlling. A reduction in the activation energy rel-
ative to that of the native enzyme, indicates that kinetics are not rate limiting.
If external resistance is rate controlling, a mass transfer coefficient, KS, can
be introduced. From the fluid dynamics of the reaction vessel, it is possible to
estimate substrate concentration at enzyme catalytic sites according to the
steady state relationship:
K,(Sf-S) = R(S) (33)
which for a Michaelian enzyme reduces to:3
K,(Sf - S) = V,,,&Y(K’NI + S)
When the internal mass transfer/reaction step is rate limiting, an effective-

ness factor, 77, is usually introduced related to dimensionless parameters char-
acteristic of the reacting system as a Thiele modulus.“e It is worthwhile noting
that most of the available correlations are based upon theoretical models assum-
ing diffusion as the only mass transfer pattern. Hence, effects related to exter-
nal mass transfer resistances are neglected.
Comprehensive models of unstirred enzyme gel flat membrane reactors have
been proposed.30~32r33~3s
The analysis of unstirred membrane systems in flat slab configuration is
carried out in two different cases:
(1) The gel formed on the flat membrane is homogenous,

(2) The gel is heterogenous, i.e., preferential flow patterns exist for the
substrate flowing under pressure through the gel.
Case I: If a homogenous enzyme gel is formed above the membrane sur-

face, no segregated regions exist: substrate permeation and conversion to prod-
ucts occur throughout the whole gel layer so that both convection and diffusion
concur in determining mass transfer through the gel. Since the reaction vessel is
unstirred even external mass transfer resistances must be taken into account.
Thus, substrate steady state mass balance equations upstream and within the
gel have to be simultaneously integrated. Expressing equations in dimension-
less form pulls out three dimensionless parameters, namely the Peclet number,
the Thiele modulus and the dimensionless Michaelis constant.
d*yl dyl
-+Pe-= 0 upstream (35)
dC’* dE’
d*-f2 dY2
-+Pe-= ** Q* /(M + r2) within the gel (36)
d(‘* d[’
B.C.1 5’ +m Yl = 1 cell inlet (37)
dY1 dy2
B.C.2 5’ = 1 -=-
Yl = Y2 ;
dS’ dc’
upstream gel surface
dY2
B.C.3’ 5’ = 0 -= 0 downstream gel surface
dS’
(Dankwaerts conditions).
In most cases, enzyme kinetic behavior can be assumed to be Michaelian.

The assumption of a semiinfinite slab for the fluid reacting volume, B.C.l,
holds that gel height, xe from Equation 19, is negligible relative to cell height.
In B.C.2 and 3, both the gel layer and the supporting membrane are assumed
to be completely permeable both to substrate and product, and enzymes are
uniformly distributed throughout the gel.
An effectiveness factor is then introduced, defined as the ratio between the
overall reaction rate and the rate one would obtain if substrate concentration
within the gel were uniform at the feed level. An overall substrate mass balance
relates this factor to the Thiele modulus, a:35
( 1 - Y2(0) )
n=(l+M)- (38)
o2
The existence of convective fluxes rules out the possibility of using cor-
relations already existing. Decreasing substrate concentration is responsible for
efficiency factors less than unity across membrane thickness. Minimum concen-
tration is, therefore, the cell output value which can be easily estimated; evalua-
tion of the dimensionless parameters is possible and correlations are not nec-
essary.
Case 2: If gel structure is assumed heterogenous, substrate convection oc-

curs through preferential pathways, while simultaneous substrate diffusion and
reaction occur in segregated regions. Minimum substrate concentration is now
attained in the core of the segregated region and cannot be readily measured.
Within a segregated region, the steady state substrate mass balance equation
can be written as:35
d*X X
-= I$,1 2 (39)
d=* (M’+x)
with the following B.C.s:
B.C.1 z= 1 x= 1 segregated region/preferential (40)

flow pattern interface
B.C.2 z=o dx/dz = 0 center of segregated region
in dimensionless form.
The modified Thiele modulus, @” is expressed as a function of the character-
istic dimension of the segregated region and external substrate concentration,3s
K E X2
0” = ( ) l/2 (41)
D -f*(O) Sf
With reference to the homogenous phase reaction rate, and to the feed sub-
strate concentration, the effectiveness factor can be estimated to be:3s
dx
dz z=l
n
=(l+M) ,I2
Q
For preferential pathway geometries other than the flat slab model, h has
to be regarded as the characteristic dimension of the segregated regions corrected
by a suitable shape factor:
As far as Michaelis-Menten enzymes are concerned, 77 vs @ diagrams have
been produced’es for various immobilized enzyme configurations. They can,
therefore, be used provided that the geometry of the segregated regions is de-
fined and that external resistances are taken into account.
Modeling of unstirred UF cells equipped with flat membranes and inter-
pretation of experimental data is then relatively simple. It is interesting to note
that transmembrane pressure plays a d.ifferent role from that which it plays in
the usual UF membrane separators. On the one hand, increasing pressures lead
to increasing permeating fluxes, thus enhancing reactor productivity; on the other
hand, high effluent flow rates strongly reduce the substrate conversion.29 More-
over, the ultrafiltering surface area per unit reactor volume is quite small.
As for membrane UF units, tubular membranes fitted in cylindrical shells
in a “tube-and-shell” configuration help in improving the performance of en-
zyme gel reactors. The ratio of filtering area to volume is an order of magnitude
higher than that for membranes in a flat slab configuration, and the flow dy
namic conditions are more easily controlled.34f36 An enzymatic gel layer can be
built up on the inner wall of tubular membranes either by filtering proteic solu-
tions in a batch mode, or flushing them along the membrane wall until the gel
layer is eventually formed. In both cases, industrial operations require the en-
zymatic reactor to be continuously operated. Feeding an axial substrate stream
to the reactor gives rise to new flow dynamic conditions. High shear stresses may
in fact develop on the gel layer surface leading to partial or total removal of the
enzyme. 34f36There is evidence of stable systems, at least under laminar flow con-
ditions.j7 Moreover, under a fluid shear field enzyme molecules can be oriented
and thereby denatured. 3e In order to develop this immobilization technique for
large scale and industrial applications, work is in progress to guarantee mechani-
cal stability of the gel layer. This goal can be achieved, first of all, by strictly
controlling the axial flow rate. The enzyme gel layer can also be protected by
applying layers of water soluble or insoluble macromolecular compounds or by
forming the gel within the porous structure of the membrane where it is less
subject to shear stresses.
Flushing the substrate solution along the enzymatic gel causes the sub-
strate to be converted to product even in the axial stream. When the enzyme
is product inhibited and the effluent from the reactor is recycled, product ac-
cumulates in the feed stream thus inhibiting gelled enzymes. High axial flow
rates may reduce conversion of substrate to product in the axial stream and
enzyme inhibition, while product conversion in the permeate remains unal-
tered at a given transmembrane pressure. 2gry1@. Under such conditions, the
axial flow rate needs to be optimized since it plays an opposite role.
Yeast invertase,30p32p33 acid phosphatase,2gr32~35~37~39urease,29 p-glucosi-
dase,29 dCMP-amino hydrolase31 and malic enzyme34t36 have been immobilized
in gel form on both flat and capillary membranes. Cellulosic and polyamide
polymers have been used as supporting membrane matrices. In all instances, im-
mobilized enzymes behave in a manner almost identical to their behavior in
homogenous solution, independent of the nature of the polymer. Neither al-
losteric nor pseudo-allosteric enzymes, proteins whose kinetic behavior is af-
fected by the presence of particular compounds in the reaction environment
(ligands), show different kinetic behavior, as they do when subjected to less
gentle immobilization procedures.31r34r36
A number of procedures have been suggested in order to improve enzyme
gel kinetic and mechanical stability.
Under ultrafiltration at suitable conditions any proteic solution can give
rise to a gel layer. Besides simple enzyme ultrafiltration, two other gel forma-
tion procedures have been proposed, namely cogelation and copolymeriza-
tion/gelation. In the first case, solutions consisting of both enzymes and high
molecular weight inert compounds are ultrafiltered through semipermeable
membranes. For example, polyalbumines (inert proteins),37t39 water soluble and
insoluble compounds,40 have been used as cogelling agents. In the second case,
before the UF step, enzymes are chemically linked to high molecular weight
inert substances by means of bridge molecules.33t4i In both cases, the enzyme
microenvironment in the gel layer is characterized by a relevant protein concen-
1
0 20 40 60
Time, hr
Figure 7.25: Effect of gel layer thickness on the stability of cogelled acid phos-
phatase at 40%. Human serum albumin (HSA) polymers: (0) 2 mg, (0) 4 mg,
(0) 8 mg, (A) 16 mg.37
tration. Enzyme kinetic stability is enhanced as protein concentration increases

(Figure 7.25);37r106 enzyme settlement is then particularly favorable. Copoly-
merized/gelled and cogelled enzymatic layers appear to be mechanically stable
over a fairly wide range of temperatures and flow rates in the laminar regime.37
Due to the covalent linking procedure, copolymerized/gelled enzymes can lose
most of their original activity. Furthermore, the deactivation rate of cogelled
enzymes appears to be less temperature-dependent than that of copolymerized/
gelled preparations. These features suggest that in many cases immobilization by
cogelation is preferred, provided that it also gives satisfactory mechanical stabil-
ity.
The flexibility of enzyme gel layer reactors is fully exploited when multi-
enzymatic reactions are to be performed. It may happen that enzymes involved
in a given transformation cannot be subjected to the same immobilization pro-
cedure. Sequential enzyme gel layers can then be built up on the surface of a
membrane in the proper sequence. Series reactions can be performed in such a
set-up; products from one enzymatic layer being fed to the following ones for
further transformation.28t41
In addition, enzyme gel reactors are inexpensive and easily controlled. De-
activated enzyme may be easily replaced.
MEMBRANE SEGREGATED ENZYME REACTORS
In these reactors, enzymes or cells are not immobilized, but only confined
to a well defined region of space. The segregation of biocatalyst in the reaction
vessel is achieved by means of an ultrafiltration or microfiltration membrane
with a suitable molecular weight cutoff. In this way, enzymes and bacterial cells
are not lost in the effluent stream, and low-molecular weight products and in-
hibitors can be removed through the membrane. Enzyme segregated membrane
reactors retaining the biocatalyst in the reaction system offer an economically
attractive alternative to the design of continuous-flow equipment.
Organic synthetic membranes, either in hollow fibers or in flat slab con-
figurations, have been extensively used in ultrafiltration units as enzyme or cell
reactors for a number of applications. Starch and maltose hydrolysis,42 sugar
production,43 cellulose hydrolysisW and steroid bioconversion4’ seem to be the
most promising industrial applications of such systems. There is also growing
interest in therapeutic applications of compartmentalized cells or microsomes
functioning as a bioartificial pancreas or extracorporeal detoxification de-
vice.s7*
The development of hollow fibers with diameters down to about 100 mi-
crons makes possible “tube-and-shell” reactors with a high surface-to-volume
ratio. Biocatalytic reactors can segregate enzymes or cells either within the hol-
low fiber lumen,46 within the shell surrounding the outer surface of the fi-
bers,44~451575g or within the porous membrane support.42~45r47~53
Segregated enzyme reactors avoid the negative aspects of immobilization
techniques such as steric hindrance and enzyme deactivation due to coupling.
Reactors with Enzymes Segregated in the Lumen of Hollow Fibers

In 1971, Rony first suggested “immobilizing” enzymes within the lumen of
hollow fiber membranes.& ” Cylindrical microcapsules” were prepared by filling
the core of hollow fibers of a suitable molecular weight cut-off with an enzyme
solution, by capillary action for instance, and then sealing both ends. Bundles of
such loaded fibers were then assembled in a “tube-and-shell” reactor configura-
tion (Figure 7.26) flushing substrate solution into the shell-side, i.e., over the
outer surface of the fibers. Substrate conversion into products can be accom-
plished if membranes are permeable to both substrate and products.
out in
Figure 7.26: “Tube-and-shell” reactor configuration.

A theoretical analysis of such enzyme membrane reactors was carried out by

Rony46 and Waterland et al. 47 In the Waterland model, which refers to asymmet-
ric hollow fibers, each fiber is assumed to be at the nodal position in an equi-
lateral triangular mesh, so that it is equidistant from six adjacent fibers.
In the absence of pressure driven fluxes through the membrane and assum-
ing a laminar flow pattern, a hexagonal free shear layer should sheath each fi-
ber; the hexagon is approximated by an equivalent circle of radius Re.
Four different regions can be distinguished in the reacting system (Figure
7.27) :
(I) The core of the fiber where enzyme solution is retained,

(2) The dense thick layer,
(3) The porous spongy part of the membrane,
(4) The shell region where substrate solution flows.
Sponge
Figure 7.27: Asymmetric hollow fiber schematic: radial cross section.47
The equations of fluid flow in region 4 are then similar to those of a liquid
film falling around a cylinder:
(43)
1 d dv
- - (r’ --J = 4k
r’ dr’
B.C.1 r’ = d/a V = 0
dv
B.C.2 r’ = r,/a -0
;i;;-
where k is a constant proportional to the Reynolds number and to axial pressure

drop. B.C.1 states that there is no slipping at the fiber-fluid boundary. B.C.2 de-
tects the shear freesurface. The solution to the set of equations is given by:
d* *
v = k (r,* - - - 2 !I!- ln(r’ (44)
a2 a*
with
3 r,* t d* 2 re2
k- -2ln (:)
2 a2 re 2 -d
The mass transport equations for substrate concentration can be expressed

as follows:
i a
--
(r (45)
r’ ar’
1 a a33
--
(r’ F) = O
r’ art (46)
i a as2
-- (r’
r’ at-’ ;) = O
(47)
Assuming a Michaelis-Menten rate equation, substrate mass transport through

region 1 is governed by:
i a as1 4 Sl
(48)
-- 0” F)
r’ art =
(0’ + 31)
S4( r’,z’) = 1 z’< 0
Z’>O
as(’
“4 r’=
r
-E ) = 0
ar’ a
S3( $, 2’) = S”(d)
as3 d,-n,as,
D3 - (r’= a (r’= $ >
ar’ ar’
y’ s2( f, 2’) = S3( p, 2’)
as2 (r’= ; ) = D3 as3 (r’= $ )
D2art ar’
Sl(1, 2’) = y’ S2(1, 2’)
asl
; (r’=l) = D2 as2 (r’=l)
D1
ar’
asl
art
(r’=O) = 0
Dimensional analysis of the problem for this reactor configuration pulls out
three dimensionless parameters, the Peclet number, Pe, 0’ and @ defined as:
kQa
Pe = (49)
TI (r* - d* ) TI
e 4
The set of the above equations leads to a nonlinear problem which can be
solved by an iterative numerical technique?’ Substrate concentration profiles, and
therefore substrate bulk concentration, are then a function of seven dimension-
less parameters ti2, z’, b/a, d/a, DJDs, DlIy’D2, 0’. Essential predictions of the
model have been estimated by Waterland et al in terms of substrate bulk concen-
tration as a function of x2, 8’, and z’, the main dimensionless parameters.47 Par-
tition coefficients were kept equal to unity and geometrical parameters were set
according to the dimensions of a typical hollow fiber, that is a = 100 pm,
b = 100.5 pm, d = 175 pm; diffusion coefficients in region 1, 3 and 4 were as-
sumed equal to each other, while the substrate diffusion coefficient in region 2
was assumed one order of magnitude lower.
A plot in shell coordinates of dimensionless length z’ vs the Thiele modulus
x” with 8’ and the degree of conversion at given values (Figure 7.28) helps in
predicting reactor performance at different operating conditions. As the Thiele
modulus increases, the reaction rate becomes diffusion controlled where the di-
mensionless length required for a given conversion is independent of x”. Under
these conditions an increase in the activity or the amount of catalyst has little
or no effect on the dimensionless length necessary for a given conversion. Ac-
cordingly, if a reactor is operated in a diffusion controlled regime enzyme ac-
tivity decays, down to 10% or even to 1% of initial activity, negligibly affect
reactor conversion. Operation of such enzyme reactors in a totally or partially
diffusion controlled regime can therefore lead to an incorrect determination of
biocatalyst stability.
IO .
I
_ 8’- 1.0)
N
I
I I I I I
-360
.391
.MI
.237
.I62
.I37
IO1
on0
.04x3
I#
10-2 10-I I IO I02 lo3
(2
Thiele Modulus, x
Figure 7.28: Reactor dimensionless length (z') as a function of the Thiele mod-
ulus. The parameter X is substrate conversion4’
As the Thiele modulusdecreases, the reaction becomes kinetically controlled.

Transition from one regime to the other and the Thiele modulus value above
which the reaction is diffusion controlled is strongly dependent on the para-
meter 6’, i.e., on the initial substrate concentration. As Sf increases, the region
under kinetic control is shifted towards larger Thiele moduli. Therefore, increases
in catalyst activity or in the amount of catalyst are effective only at high sub-
strate feed concentrations.4749
Plots of substrate bulk concentration vs the dimensionless reactor length4’
show how steep the descent of profiles can be when large Thiele moduli are ap-
proached. Substrate conversion is therefore more rapid in the diffusion con-
trolled regime, and reactors are better operated in this regime.
Furthermore, in applications where products can be easily separated from
the effluent stream, it can be convenient to operate at small dimensionless lengths
at the expense of lower conversions. Product mass flow rate will increase, more
than compensating for this reduction in conversion.
Evaluation of stability and catalytic properties of the immobilized system
must take into account possible pH differences between the inner core of the
fiber, where the reaction takes place, and the bulk of the feed solution. The
production of compounds which alter the pH, like ammonia produced from
urea via immobilized urease,48 and the partition properties in hollow fiber mem-
branes can result in creating such pH gradients. Experimentally, these differences
produce more or less pronounced shifts in the optimum pH dependence of en-
zyme activity relative to its free form dependence and thereby affect the ac-
tivity of the enzyme at work.2~3~~
Urease, uricase,48 alcohol dehydrogenase,” and alkaline phosphatasesl have
been used in such reactors.
Given the availability of hollow fiber membranes adequately permeable to
substrates and products, and the control of fluid flow all around the fibers in
the bundle in order to assure uniform flow distribution and to avoid stagnation
(in order to reduce mass transfer diffusional resistances), the technique offers
several advantages. Enzyme proteins can be easily retained within the core of the
fibers with no deactivation due to coupling agents or to shear stresses, and the
enzyme solution can be easily recovered and/or recycled.
Similar reactor configurations using flat membranes in place of hollow fibers
have even been used with urease, uricase, glucose oxidase and creatinine kinase.49
An enzyme solution is, in this case, introduced into one of the membrane-sepa-
rated chambers of a flat-membrane dialyzer. Operating the reactor at high feed
flow rates leads to a reduction of mass transfer resistances, which can also be
achieved using suitable turbulence promoters. The overall behavior of such sys-
tems appears to be similar to that of the hollow fiber enzyme reactors, except
for an apparent higher efficiency.49
Reactors with Enzymes within the Pores of Asymmetric Membranes

Asymmetric synthetic hollow fiber membranes designed for use in ultra-
filtration/dialysis processes can provide an interesting support for immobilizing
enzymes.
The term asymmetric refers to membranes comprised of a porous spongy
wall supporting a very thin dense layer. The thin skin layer, approximately 0.5
I.tm thick, with pore sizes in the range of IO to 2000 a, determines the separa-
tion properties of the membrane with little hydraulic resistance to mass trans-
port. This skin layer is fully surrounded by a spongy structure, approximately
75 Pm thick, with pore sizes of 5 to 10 Pm, and 80 to 90% porosity which has
a very high hydraulic permeability. This porous spongy wall acts as a mechan-
ical support for the semipermeable thin layer.
The use of these fibers in bioreactors can eliminate some of the disadvan-
tages of the previously discussed reactor systems.42t4T.s3
Enzymes can be entrapped within the outer sponge layer of the fibers by
saturating their pores with an enzyme solution (Figure 7.29). The wet catalytic
fibers can then be organized in bundles and assembled in a “tube-and-shell” re-
actor configuration. If the pores in the dense layer are small enough to retain
enzyme molecules but large enough to freely pass substrates and products, the
enzymes are effectively immobilized or segregated within the spongy annular
section. The reactor can then be operated by feeding substrate solution through
the lumen of the fibers, in a reverse configuration than that previously sketched.
If substrate molecules are small relative to the membrane’s molecular weight
cut-off, they diffuse across the dense layer from the fiber lumen into the enzyme
solution where the reaction takes place. In turn, product molecules diffuse back
into the substrate solution, or, if gaseous, leave the fiber through the shell side.
Figure 7.29: Schematic of the cross section of a hollow fiber in whose macro-
pores enzymes are immobilized.42
The dynamics of substrate conversion therefore depend on enzyme kinetics

as well as on mass transport conditions. Diffusion through the membrane matrix
and within the flowing solution, play the most important roles in transport
mechanisms. Since the flow is laminar in most cases, substrate and product
transport resistances through the dense layer are exceedingly small relative to
diffusional resistances in the flowing solution. The rate-limiting step in sub-
strate conversion is therefore either simple diffusion or the intrinsic kinetics of

the reaction itself.
Mathematical modeling of such reactors has been extensively investi-
gated. 47f53 Referring to the model presented in the previous section (Figure
7.271, the reaction system may be divided into only three regions; namely the
previously defined regions 1, 2, and 3, with substrate solution flowing in region
1. It is assumed we have:
0 Laminar flow in region 1, with substrates and products diffusing

through all three regions
0 Steady state conditions
l Chemical reaction confined to region 3, where fluid is assumed to
be stagnant
l Constant fluid properties.
Using a set of cylindrical coordinates as in Figure 7.30, the set of mass trans-
fer differential equations defining substrate concentration in each region is:
i a
D3 -- (r as,) =R (50)
r ar ar
1 a
as2
(51)
'D2 - - (r - ) = O
r ar ar
1 a
-~ asl)_EL (52)
D1 (r
r ar ar ax
Assuming a laminar flow pattern, the radial velocity profile in region 1 can
be expressed as:
r2
v(r) = vmax (1 - - ) (53)
a2
with enzyme kinetic behavior still modeled by the Michaelis-Menten rate equa-
tion.
R = Vmax S3
K’M + s3
----- x
Ultrothin
Membrane
Figure 7.30: Axial section of an asymmetric hollow fiber. The substrate solu-
tion is fed to the lumen of the membrane under laminar regime.47
The intrinsic nonlinearity of this set of equations does not permit an ana-
lytical solution. Lewis et aIs and Davis”’ have proposed an analytical solution
to the problem in the case of low feed substrate concentration, that is for a
linear rate equation. An iterative numerical solution for the nonlinear problem
was fully developed by Waterland et al.47
Dimensional analysis of the equations relates the concentration profile, both
along the lumen and the radius of the fiber, to seven dimensionless parameters:
Thiele modulus, X12, dimensionless length, z’, dimensionless Michaelis constant,
6’ and the following dimensionless quantities: b/a, d/a, Dr/Ds and DI/i’D2.
General correlations of bulk concentration as a function of Thieie modulus,
dimensionless Michaelis constant and the dimensionless length in the entire range
of practical operating conditions are available under the same assumptions as the
previously described reactor configuration.47r53
Plots of the dimensionless length required to achieve a certain given conver-
sion as a function of the Thiele modulus, with the dimensionless Michaelis con-
stant as a parameter, are qualitatively in agreement with those obtained with the
reactor operated in the reverse configuration. Regions under diffusion or kinetic
control are approached as the Thiele modulus is increased or decreased, respec-
tively (Figure 7.31). When the dimensionless Michaelis constant 8’ (i.e.,the recip-
rocal of the feed concentration) is decreased, transition between the kinetic and
diffusion controlled regimes occurs over a narrower range of the Thiele modulus.
Kinetic control holds for larger Thiele moduli, and the slope of thelinear depend-
ence of the dimensionless length z’ on the Thiele modulus (in a log-log plot) in-
creases. An increase in feed substrate concentration may therefore shift reactor
operation from a mainly diffusion controlled regime to a mainly kinetic con-
trolled regime, with a consequent decrease in reactor conversion. From another
point of view, one might say that increases in feed concentration require sensible
increases in the reactor dimensionless length to achieve the same conversion as
before the change.
Comparisons between the performance of the reactor operated in this con-
figuration and in the reverse one evidence 47 that both behave in a similar way
%*
,266
161
.0677.
0440
.0259
0126
10-I I IO IO’
Thiele Modulus, ?
*lu !- -:--
f’
f$
3
::
-u
(
‘----t
.-
-.-
-. -.I_
I-
_...
s ,o-’
g I-
.i_--
, i-_
:
z
a
42 - 00.50
Thicle Modulus, A
IOr-rl-l-l---l-r-r--r~n
Thlele Modulus, P
Figure 7.31: Reactor dimensionless length (z’) as a function of the Thiele mod-
ulus at various dimensionless times. The parameter X is substrate conversion.47
under conditions of kinetic control. When the Thiele modulus is increased and
the enzymes loaded into the sponge of the hollow fibers, the dimensionless re-
actor length required to achieve a given conversion is smaller than that nec-
essary for a reactor where enzymes are encapsulated within the lumen of the fi-
bers themselves. Reactors with enzymes entrapped in the macroporous region
of the membrane wall appear to be better whenever they are operated in the
diffusion controlled regime.
Experimental work in which enzymes with relatively simple kinetics were
immobilized in the sponge of hollow fibers, e.g., a-galactosidase,54 invertase,54
glucose isomerase,55 and urease,53 are in good agreement with predictions of the
theoretical model. 47~53Quick graphical procedures are available in the literature
for evaluating the extent to which external and internal diffusion affect immobi-
lized enzyme kinetics. 43 Agreement between experimental results and predic-
tions of the more comprehensive proposed model suggeststhat it might be used
to predict reactor performances in the case of simple kinetics and when the
kinetic and transport parameters are known.
When complex kinetics are involved in substrate conversion, as with prod-
uct-inhibited enzymes (like amyloglucosidase which catalyzes maltose conver-
sion to glucose4*) or with reactions involving a number of intermediates (like
starch hydrolysis by means of amyloglucosidase4*) definite information on en-
zyme kinetics is rarely available. Moreover, when solutions of macromolecular
compounds, like starch, are fed to the reactor, diffusional resistances are more
pronounced and may hinder the possibility of using such reactors for analytical
purposes. A thorough knowledge of the chemical mechanisms through which
substrates are converted to products and of the coupling of transport phenom-
ena to enzyme reactions appears to be a prerequisite for the design of such bio-
chemical reactors.
Different operating conditions may require some modification of the per-
formed analysis. Ultrafiltration and/or osmosis can promote convective solute
or water flux through the membrane wall. Should it happen, radial convection
could compete with diffusion as the main substrate and product transport mech-
anism. The relative importance of the two transport mechanisms can be evalu-
ated by comparing the radial convective velocity to the diffusive velocity, that
is the ratio of the diffusion coefficient to the wall thickness. When the first
one is negligible relative to the latter, the model applies without modification.
The second possible effect of the radial flux is to remove enzymes from the
fiber wall, resulting in the reduction of reactor efficiency.
This reactor configuration is particularly attractive since substrates are phys-
ically separated from the enzyme solution only by a very thin membrane layer,
the dense skin, thus minimizing mass transfer diffusional resistances. A number
of other advantages make this reactor configuration feasible for many applica-
tions. The use of small diameter fibers leads to large area-to-volume ratios, with
high enzyme loading capacity per unit of reactor volume. The enzyme micro-
environment is fully shear free. Moreover, membranes act as selective barriers
protecting enzymes from macromolecular contaminants such as proteolytic en-
zymes, or selecting substrates on the basis of their permeability or electric charge.
Many enzymes can even be coimmobilized with the macroporous region of asym-
metric membranes. The procedure itself does not exclude the possibility of cross-
linking enzymes directly to the porous polymeric matrix of the fibers, nor that
of compartmentalizing within the same region enzymes previously coupled to

soluble polymers or inactive proteins. Such binding procedures might in fact
result in a greater stability, that is a slower decay of enzyme activity with time.
In this reactor configuration, once products are formed they diffuse back
towards the stream flowing in the core of the fibers. Such configuration is there-
fore feasible for all those applications in which products at relatively high con-
centration are tolerated in the circulating stream.
Tube and Shell Membrane Reactors with the Biocatalyst on the Shell Side
As with enzymes, whole active microorganisms can be segregated in a def-

inite region of space by means of membranes in order to catalyze specific reac-
tions. Microsomess7& and bacteria 61 have been and are currently employed in
membrane reactors in order to perform complex multienzymatic reactions or to
reduce overall reactor costs, avoiding enzyme purification. When the size of the
biocatalyst exceeds the dimensions of the pores in the sponge of the asymmetric
membranes,44 previously examined reactor configurations are not applicable.
Nevertheless, the problem can be overcome allowing the biocatalyst to stay on
the shell side, while the substrate solution is kept flowing within the lumen of
the fibers.
Apparently the main transport mechanisms through which substrate conver-
sion takes place are:
0 Diffusion of substrates from the bulk fluid phase to the membrane

l Diffusion of substrates within membrane pores
0 Diffusion of substrates within the shell-side of the reactor to the
biocatalyst
Referring to one fiber alone, the scheme of the reacting system is similar to
those examined so far. However, the reaction does not occur in either regions 1,
2 or 3. The set of differential mass transfer and continuity equations defining
substrate and product concentration in these regions are equal to those pre-
viously examined. The description of mass transport in the shell-side region is
somewhat more complicated. Differences in the environment surrounding each
fiber, the position of fibers in the bundle, and the ultrafiltration fluxes make
both the analytical and the numerical approach quite difficult.
Models proposeds7~58~62 ‘often deal with simpler systems, assuming good
mixing conditions both in the shell and within the lumen of the fiber, and pure
diffusive fluxes. In some particular configurations, as in the artificial pancreas,
such assumptions hold and the models work quite well, especially in the case of
membrane units using one large hollow fiber alone.57f63 In cases where a quick
transient response is needed, the only way to circumvent the slow transient
behavior of the device is to reduce the volume where the catalyst is compart-
mentalized.57 Even though diffusion appears to play an important role in
substrate and product transport, there is experimental evicence that the bun-
dles of hollow fibers assembled in a “tube-and-shell” configuration respond
more quickly than could be predicted by assuming purely diffusive fluxes across
the membrane walls.62 Pressure drop along the length of each fiber should there-
fore produce a transmembrane pressure across the membrane wall such that at
the inlet of the reactor, the pressure difference promotes an ultrafiltration flux
towards the shell side, where the catalyst is, and at the outlet it promotes a back-
ward flux from the shell towards the lumen of the fiber (there is only a small
pressure drop in the shel144). These pressure profiles promote fluxes which im-
prove the performances of the system as compared to those exhibited by pure
diffusive reactors.
A quantitative analysis of such ultrafiltration flux59J62 can be approached
referring to a single fiber device (Figure 7.32). At a distance, x, from the inlet
of the reactor local ultrafiltration flux, J, can be related to local transmembrane
pressure, AII, according to the following equation:
J(x) = L Ag (x1 (54)

P
where the coefficient Lp is the hydraulic permeability of the membrane to the
substrate solution. In turn, the local transmembrane pressure can be expressed
as:
ANx) - rib(x)) - flonc - n, (55)
where TIonc and I’& are the local oncotic pressure and shell-side pressure. As-
suming that the feed solution is newtonian with a laminar flow pattern, pressure
in the fiber at distance x from the fiber inlet can be evaluated from Poiseuille’s
law to be:
8PQ
np = nIbi - ( 1 x (56)
n a4 N
f
where flbi = inlet pressure
c1 = substrate solution viscosity
Nf = the number of fibers in the bundle.
Introducing 8 = gbf _ IIcnc - IIs
8lJQ
and C -_
n a4 N
f
and substituting the expression for flux as a function of transmembrane pres-
sure one obtains:
J(x) = La (6 - Cx) (57)
A decrease in hydrostatic pressure along the fiber due to resistance to sub-

strate solution flow occurs so that at a definite distance from the inlet, say Lc,
transmembrane pressure is nil. Fiber-to-shell solution flux from that point on is
negative and becomes a shell-to-fiber flux. Neglecting the shell pressure drop, the
overall fiber-to-shell ultrafiltration net flow rate can then be obtained upon in-
tegration of the flux equation over the length of the fiber from the inlet to Lc,
that is:
L
= 2 II a Nf LP ‘(B - C x) dx =
Qfl I
0 (58)
L2
=2aaN L (BL -C-‘)
C
f P 2
But at a distance Lc from the inlet ALI = 0, so that Lc = B/C and conse-
quently:
B2
=naN L - (59)
Qfl f PC
The overall shell-to-fiber flux can be obtained in a similar way, upon integra-
tion of the flux equation between Lc and the reactor outlet, as follows:
L-Lc
dx = - n a Nf LP (L - ~c)* c
Qf2 =
- I
2 ‘IIaNfTAP
0
C x (60)
The shell volume is constant, therefore Qfl + 9f8 = 0 (61)
or in other terms II a Nf L,, B* /C = II a Nf LP (L - Lc)* C
that is 8*/C = (L - Lc)* C and L = 2 B/C = 2 Lc
The expression for B can be substituted in the expression of the overall fi-
ber-to-shell flux to give:
Qfl = 2 P LP L* Q/ a3 (62)
Upon imposition of the pressure drop along the fibers, the flow rate Q can
be evaluated from Poiseuille’s law to give a relation where the net ultrafiltration
flow rate, QfI, is related to the total membrane area, A, and pressure drop along
the fibers:
gfl = A LP AIT/ (63)
The characteristic time of this reacting system is 7 = V/QfI, where V is the

shell-side void volume. The previous equation can be rewritten to relate the char-
acteristic response time of the device and the volume where the biocatalyst is
confined to give:
V = A LP All r /8 (64)
Given the membrane surface area, A, its hydraulic permeability, Lp, and
pressure drop along a fiber, AII, it is possible to estimate the minimum volume
required in order to obtain the necessary response time.s9
This reactor configuration is often appropriate for complex catalytic sys-
tems. The use of cofactors with the enzymes in continuous flow systems has of-
Figure 7.32: Schematic of a hollow fiber encased in a closed shell. The occur-
ring absorption and reabsorption fluxes along membrane axis are QfI/Qf2, re-
spectively. No transmembrane flux occurs at membrane half length.59
ten been limited by the need to supply large amounts of fresh cofactor, usually
an expensive compound. Many procedures have been suggested in order to con-
fine these low-molecular-weight compounds in a well defined region of space
where they are continuously used and regenerated.103r111When the biocatalyst
is compartmentalized in this way, cofactor costs are reduced. In the presence of
a suitable regeneration system, only low cofactor additions are needed to main-
tain excess concentration levels thus assuring maximum rates of conversion.60j64.
Membranes which retain high-molecular-weight biocatalysts in the shell, but
allow the passageof low-molecular-weight products or inhibitors, may also allow
the segregation of biocatalysts in the shell capable of transforming inhibitors
into nonactive compounds, thereby keeping their concentration low in the ef-
fluent reacting stream. This concept has been used in the enzymatic hydrolysis
of celIulose.44~65A blend of cellulose, substrate, and cellulase, a suitable enzyme
mixture, is kept flowing within the lumen of the fibers assembled in the bundle.
When cellulose is converted to glucose, cellobiose (an intermediate product) is
formed which acts as an inhibitor for some cellulase enzymes. The presence of a
suitable biocatalyst, such as the enzyme/3-glucosidase440r the cells of Hansenula,65
in the shell of the reactor can reduce the extent of inhibition. As soon as cello-
biose is produced, it flows towards the shell being converted to glucose by fl-
glucosidase. In this way, the inhibitor concentration in the reacting stream can
be kept low without affecting the overall sugar concentration.44
Reactors in this configuration are also employed in therapeutic applica-
tions. Bundles of hollow fibers59r66or a single large hollow fiber57r63 in a cylin-
drical module are used to separate blood flowing within the lumen and mammal-
ian pancreatic islets as assistance to diabetic patients. They have also been sug
gesteda for use as extracorporeal blood detoxifiers, where mammalian liver mi-
crosomes are compartmentalized in the shell of the module. In such systems, if

the membrane molecular cut-off is carefully chosen, membranes will protect
transplanted cells from the immunodefensive action of leukocytes in the blood.
The main drawback of membrane reactors in this configuration is the relatively
slow response to metabolic stimulation due to large shell volumes required to
accommodate a number of large-sized biocatalytic units needed to perform the
assistance action.
Diffusion and ultrafiltration fluxes due to pressure drop along the length of
fibers play the most important role in substrate and product mass transfer when
systems are operated as previously described. However, an ultrafiltration flux
can be promoted from the lumen of the fibers outwards and/or from the shell
inwards.44 Better reactor performances should result from such operating con-
ditions.
Perspectives
Most of the membrane segregated enzyme systems previously examined suf-
fer some constitutive drawbacks which limit their yield and area of application.
When enzymes are entrapped within the sponge of asymmetric membranes,
product and substrate mass transfer occur mainly by a diffusive mechanism;
reactor performance is then controlled only by means of the amount and kind
of charged enzyme, and the fluid dynamics of the solution in the core of the
fibers. UF or RO fluxes, moreover, result in enzyme losses. Enzyme crosslinking
in the membrane pores can reduce these losses, but it can lead to an initial activ-
ity loss, as compared to that of the native enzyme. Of course, once the enzyme
is deactivated, it makes the reactor useless for further operation. Such immobil-
ization techniques are seldom useful for microbial cells due to their large size.
In recent years, membranes have been cast with microbial cells incorporated
in the casting solution. To date, the casting of UF and RO membranes charged
with enzymes or whole cells on an industrial scale has been limited by the dras-
tic conditions under which the synthetic membranes are usually formed. The
presence of nonaqueous solvents in casting solutions, and the high temperatures
required by membrane annealing, generally denature enzymatic proteins, with a
loss in their catalytic activity. Recently, a number of microorganisms, e.g., Sulfo-
lobus solfataricus, have been discovered which can withstand both high tempera-
tures and organic solvents. Microbial enzymes maintain their activity in condi-
tions otherwise denaturating, presumably because of cellular membrane protec-
tion. 61 Polysulfone 61r67and cellulose acetate membranes68 have been cast with
microbial cells in the casting solution using the phase inversion technique, as well
as in polyurethane foams.6g Cell, loaded membranes appear to be kinetically ac-
tive and stable over a long period of time;67 the resistance to flow is fairly low
and is strongly dependent on the forming procedure. It is noteworthy that cell
entrapment can enhance microbial activity as compared to cell behavior in ho-
mogenous solution,61 an effect probably due to cellular membrane permeabiliza-
tion as a consequence of the entrapment procedure.
Entrapped whole cells are the source of a number of microbial enzymes use-
ful for industrial purposes. In addition, the possibility of long term operation
make this immobilization procedure extremely attractive. Stable immobilized
cells permit the carrying out of an enzyme reaction in one stage alone, an op-
eration which normally requires two different stages, namely enzymatic con-
version and membrane separation.
MEMBRANE BOUND ENZYMES IN CONTINUOUS-FLOW SYSTEMS
In recent years a large number of techniques have been suggested in the Iit-
erature for immobilizing enzymes on insoluble carriers; in particular, immobiliz-
ation on polymeric solid supports or glass beads. A stable attachment can re-
sult from ionic binding, cross-linking and covalent linking to a water-insoluble
matrix. A full description of chemical and/or physical procedures required to
make enzymes insoluble is beyond the scope of this chapter and is extensively
dealt with elsewhere.112~113
We will be concerned mainly with enzymes bound to synthetic polymeric
membranes via covalent binding. Since 1954, when protease was covalently
bound to diazotized polystyrene,“4 enzyme immobilization via covalent bonds
has been an established immobilization technique, usually carried out by means
of extremely active bridge molecules, such as CNBr, or hi/multi-functional rea-
gents, such as glutaraldehyde. ‘13 However, the mechanisms involved in enzyme
immobilization are not well understood in most cases. When glutaraldehyde is
used as a coupling agent, it has been suggested that immobilization results from
reaction between the enzyme free amino groups and the glutaraldehyde alde-
hyde functions with the formation of an intermediate Schiff base. The resul-
tant bonds are generally extremely stable, due to the high binding energy of the
covalent bonds. On the other hand, coupling agent molecules, usually quite
small, can penetrate deep into the active sites of enzyme protein coils where re-
action takes place. Once these sites are involved in linking to the matrix, they
are no longer available to substrate molecules, resulting in a irreversible loss of
activity as compared to the initial activity of the native enzymes. When the ex-
tent of initial denaturation is acceptable in the economics of the process, en-
zymes bound to membranes can be used in continuous flow reactors. Aspara-
ginase, arginase, ligandine and glutathione-S-transferase,” catalase, uricase, allan-
toinase and allantoicase,71 have all been immobilized onto Cuprophan hollow fi-
ber membranes; asparaginase and alcohol dehydrogenase onto nylon tubes;72j73
and urease, dextranase74 and papain” have been attached to cellulosic and poly-
sulfone flat membranes. Apparently the immobilization procedure affects the
immobilized enzyme activity more than the membrane configuration.
Enzymes are covalently immobilized primarily onto the surface of the mem-
brane exposed to the feed solution, known as the “active side” of the asym-
metric membrane. In general, it is not clear whether reaction between enzymes
and polymeric membranes via coupling agents simply results in enzyme attach-
ment to the membrane, or if it leads to an enzyme-carrier network inside the
polymer matrix. For the sake of simplicity let us assume that asymmetric mem-
branes are used, that suitable active groups are available on the polymeric sur-
face and that the membrane molecular weight cut-off is such that the active
layer is enzyme-impermeable. In this way, even though their activity is often
drastically reduced, surface bound enzymes are in close proximity to the sub-
strate solution-thus reducing the mass transfer resistance to that associated
with the boundary layer. When enzymes are covalently immobilized in the
sponge of the membranes, the mass transfer through the membrane wall must
also be taken into account.
Applications of covalently immobilized systems are essentially:
(1) Membrane electrodes for analytical purposes;
(2) Reactions of substrates whose molecular weight is low as compared

to membrane molecular weight cut-off;
(3) Enzymatic conversion of macromolecules to lower molecular weight

species able to permeate the supporting membrane.
As far as membrane electrodes are concerned, readers are referred to the
When the substrate has a molecular weight lower than the membrane cut-
off, enzymes can be covalently bound either to the “active side” of the mem-
brane or within the sponge-like substructure of the membrane.
If the enzyme kinetic behavior is not affected by compounds in the solution

to be processed, enzymes are preferentially bound onto the “active side” of the
supporting membrane, thus minimizing the overall substrate mass transfer re-
sistance. When such systems are used for analytical purposes, the diffusional re-
sistance in the bulk phase must be taken into account. Obviously, the kinetic
parameters for native enzymes are no longer applicable for the immobilzed-
system. 1.-3
Comprehensive descriptions of mass transfer and kinetic effects on the per-
formance of such reactors under different operating conditions are not yet avail-
able. A theoretical analysis of a tubular reactor with impermeable inner walls
coated with enzymes was carried out by Kobayashi and Laidler76r77 and ex-
perimentally confirmed by Bunting and Laidler.73 In this analysis, the reacting
solution is fed to the core of the fibers. Assuming laminar flow, the steady state
mass transport equation for the substrate is:
2 i as r2 as
D (Z+-- Vmax (1 - -) - = 0 (65)
r ar) - a2 ax
B.C. 1 x=0 s = Sf
B.C. 2 r=a D; = R(SW)
B.C. 3 r=O s = Sf
where Sw is substrate concentration at the membrane wall.

Analytical solutions have been obtained assuming the enzymatic reaction as
the controlling step which implies flat substrate concentration profiles all along
the tube. Numerical solutions are also available both for diffusion controlled
mass transport, when the substrate concentration at the wall can be assumed to
be nil; and in the so-called transition region, when the reaction rate is neither ki-
netically nor diffusion controlled. They describe the dependence of the overall
reaction rate on substrate concentration, the Michaelis constant, and the Damk-
holer number (the ratio of the reaction rate in the absence of diffusional resis-
tance to the diffusion controlled rate).
This model does not apply to a porous wall tubular reactor. In this case,
we must account for solution losses along the tube length, and a radial convec-
tive term must be included. Again, the enzymatic reaction at the catalytic wall
enters into the model as a boundary condition.
In addition to applications in industrial processes, enzymes bound to the
active side of hollow fibers assembled in a “tube-and-shell” configuration7’ have
been and are under study as extracorporeal or “in viva” devices for use in he-
patic failure or to assist leukemic patients. For these therapeutic applications,
the irreversible binding of the enzyme to the membrane is extremely important.
Enzymes are in fact purified from mammalian, non-human sources, and they are
continuously in contact with circulating blood; their removal from the solid
matrix would result in a series of immunodefensive reactions which would alter
liver operating parameters. Enzyme immobilization via covalent binding gen-
erally meets this requirement. Hollow fiber membrane reactors with covalently
bound enzymes have been proposed as extracorporeal blood detoxifiers or as
devices to reduce arginine and asparagine content in the blood of leukemic pa-
tients. A group of enzymes from pig liver cytosol, known as glutathione-S-trans-
ferase, have been immobilized onto the active side of commercial hemodia-
lyzers in order to reduce the toxin content in blood. These enzymes change the
polarity of the toxins through conjugation to glutathione. The resulting shift
from hydrophobic to hydrophilic facilitates release through the membrane wall.
Arginine and asparagine content in the blood of leukemic patients has been re-
cently related to neoplastic cells growth; therefore their removal should hinder
further cell development.‘s Asparaginase and arginase79 have been immobilized
in commercial Spiraflow hemofilters in order to support anti-leukemia therapies.
Ex vivo and in vivo experiments circulating blood through these devices gen-
erally confirm an effective decrease in blood arginine and asparagine content
after a relatively short time of extracorporeal circulation treatment.70t79r80
When symmetric membranes are used or when enzymes are fed to the
spongy part of asymmetric membranes, enzyme immobilization results in either
a uniform fixation of enzymes throughout the membrane wall, or in the forma-
tion of a carrier-enzyme insoluble network in the sponge of the membrane. Mass
transfer through this solid phase must therefore be taken into account. A theo-
retical model neglecting radial convective transport and the dense layer in asym-
metric membranes is available in the literature.81 The reacting solution is still as-
sumed to be fed to the core of the hollow fibers. Steady state, laminar flow, and
isothermal conditions are assumed. Moreover, the enzymes are assumed to be
uniformly distributed and the membrane wail curvature is neglected. Differen-
tial dimensionless mass balance equations can be written as follows:
a+- 1 a
as,
L
(66)
i (1 - rJ2) ; - ;, $r’ ar,
in the bulk liquid phase,
I# 1 sS
-_- = 0 O< r”<l (67)
arl~2 r* 1 + B* ss
in the annular catalyst layer, with the following boundary conditions:
a&
-_=O r' = 0
ar'
sL = ss r' = 1; r" = 0
a.+ ass
o----z- r’ = 1; r” = 0
ar' ar"
ass
-_=O r" = 1
ar"
SL = 1 2' = 0
The concentration history appears to be a function of three dimensionless

parameters, a modified Thiele modulus, y*, the mass Biot number, w, and the
dimensionless feed concentration, 0”. The set of non-linear equations is un-
coupled by introducing an effectiveness factor, Q, and numerically solved. In
order to reduce computer time, the effectiveness factor has been conveniently
expressed as a weighted sum of its value for the zero and first order reaction
rate. Different regime conditions are depicted in terms of a dimensionless para-
meter,
(r* “)_1 = "max (d-a) , 2
K'M a
analogous to the Damkholer number. This is called the reactor modulus and is
obtained by collecting the mass Biot number and the Thiele modulus together.
The progress of the reaction is described in terms of the reactor modulus,
(Y”WP, and the dimensionless reactor length, now Z’ = DL x /2 vmax a’, the
latter being equivalent to the inverse Graetz number and proportional to the re-
actor residence time (see Figure 7.33). Results resemble those obtained by Water-
land et al in their model of hollow fiber reactors with the enzyme compartmen-
talized in the sponge of the polymer matrixP’ At a low dimensionless reactor
length, and a high reactor modulus, the reaction is diffusion-controlled; at low
reactor moduli, the reaction is kinetically controlled. As in the model proposed
by Waterland et al, the transition region becomes narrower as the feed substrate
concentration increases. Dimensionless reactor lengths, Zkin and Zdiff, define
conditions to achieve a given conversion under diffusion and kinetic control.
-2L I I I 1 I I I I I
2
r3
AI
8
i-0
-I
-2
-I - p&N
-2 I I
-3 -2 -I 0 I
log 2’
Figure 7.33: Reactor modulus as a function of reactor dimensionless length at

various values of the reciprocal of the dimensionless Michaelis constant. The
parameter X is substrate conversion.81
Generally the dimensionless reactor length required to achieve a given conver-

sion can be expressed by the approximate relation:
where <v> is an average catalyst effectiveness factor. This expression simplifies

the treatment of data and reactor design considerably.
Horvath et aIS1 also give suggestions for the estimate of a reactor effective-
ness factor using dimensionless reactor lengths under limiting kinetic and diffu-
sion controlled regimes. The reactor effectiveness factor can be expressed as the
ratio of the actual reaction rate to a maximum rate characteristic of the reacting
system, at a given conversion. The average reaction rate is inversely proportional
to the dimensionless length; the dimensionless reactor length at limiting condi-
tions can therefore be used to define an overall reactor effectiveness factor as
the ratio of a minimum length characteristic for the reactor to the actual. Re-
actor efficiency can thus be related to a kinetic efficiency, Ekie = Zkie/Z’, and a
diffusive efficiency, Ediff = &jiff/Z’. Ekin compares the actual heterogenous cata-
lytic rate to that achievable with the same amount of enzyme in a homogenous
plug flow reactor under the same conditions; it is therefore a measure of the de-
gree of utilization of the enzyme. Eoiff is the ratio of the actual rate to the rate
that would produce zero substrate concentration at the catalytic surface. Plots
of Ekin vs the reactor modulus (see Figure 7.34) show how the functional de-
pendence of f$,., on the reactor modulus is similar to that of the catalyst effi-
ciency on the Thiele modulus.
-2 0 2 4 6
-log +I
Figure 7.34: Kinetic efficiency as a function of reactor modulus at two values

of substrate conversion. The parameter is the reciprocal of the dimensionless
Michaelis constant.81
When such reactors are used for analytical purposes, the usual Eadie or Hof-
stee plots should not be used.81 However, at low reactor moduli (kinetic con-
trol), straight lines are obtained and these plots are reliable for evaluating en-
zyme kinetic parameters. In the diffusion controlled regime, due to axial changes
in the boundary layer resistance, no initial rate can be determined in the conven-
tional sense even when the catalyst effectiveness factor is unity. The highest re-
action rates are, in any case, obtained when the reaction is rate-limited by sub-
strate transport in the bulk liquid phase. Design equations can then be obtained
by the definition of limiting reactor length under bulk diffusion control. When
maximum utilization of the available catalyst is the main goal, relatively inac-
tive enzymes are needed. The reactor dimensionless length required to achieve a
given conversion is then given by the expression for Z under kinetic control:
*
f 0
zkin = - (/!I*x - ln(l-X))
a
When radial convective fluxes due to transmembrane pressure occur, this
model is no longer accurate.
Enzyme immobilization in the sponge of polymeric asymmetric and sym-
metric membranes has the advantage of a stable immobilized enzyme system
along with the improved isolation of enzymatic proteins from immunodefensive
system actions, high molecular weight inhibitors and proteolytic enzymes. A
suitable choice of membranes based on their separation properties allows sub-
strate molecules to permeate through the membrane while, at the same time,
separating undesirable compounds from the enzymatic proteins.
The mass transfer mechanisms operative in substrate conversion are essen-
tially those described by Waterland et al47 in their model of the compartmental-
ized enzyme membrane reactor. Since kinetic parameters cannot be assumed
equal to those of native enzymes, a kinetic analysis has to be performed in order
to characterize enzyme behavior after the immobilization procedure.
Sometimes even membrane transport and mechanical properties are affected
by the harsh chemical treatments required in the immobilization procedure.
Figure 7.35 shows how immobilization of urease by diazotization on hetero-
geneous polysulfone flat membranes reduces the membrane hydraulic permea-
o 74 It also shows how the higher permeability membranes are more
bility by 50/o.
affected by the immobilization procedure. After the enzyme is immobilized, it is
wise to check the integrity of the membrane (e.g., the permeability to the spe-
cies of interest and the strength).71R74
In extracorporeal devices, commercial Cuprophan hollow fiber membranes7’
are often used. In order to reduce the side effects of free enzymes in intraven-
ous injection therapies, asparaginase, catalase, uricase, allantoicase and allantoinase
have been immobilized by means of glutaraldehyde or CNBr onto these mem-
branes. In vivo fluid dynamic conditions strongly limit the range of conditions at
which such reactors can be operated. Reactors are often forced to operate in the
diffusion-controlled regime. Under these operating conditions, the overall reac-
tion rate depends on the substrate supply rate; the apparent enzyme activity can
thus increase by increasing the recirculation flow rate. Mazzola et al’l showed
that a value of the axial flow rate exists at which the maximum reaction rate is
attained; higher flow rates result in a decrease in apparent enzyme activity. In in
0 1 2 3 I 5
Pressure (bar)
Figure 7.35: Volume flow as a function of pressure using two different types
of membranes, type 2 being more permeable than type 3. Blank symbols: mem-
brane type 2. Filled symbols: membrane type 3. Test conditions: uncoupled
state, pure water (0); uncoupled state, dextran T4 (4000-6000 daltons) 1 wt%
(a); coupled with urease, using a 0.015 M urea solution (o).74
vivo experiments, long reactors may not attain maximum reaction rates, due to
high pressure drops arising from blood viscosity. Higher apparent reaction rates
result when operating with shorter reactors at higher flow rates but under the
same pressure drop. In vivo extracorporeal circulation experiments on rats with
asparaginase bound to the outer surface of Cuprophan hollow membranes assem-
bled in a “tube-and-shell” reactor configuration confirm that even this kind of
reactor can eliminate asparagine after a short period of time, i.e., 3 to 4 hours
(Figure 7.36).‘r While asparagine depletion in plasma is accomplished quickly, in
blood cells it proceeds more slowly. Unfortunately, after the treatment, the as-
paragine content in blood cells suddenly rises, and when it levels off at its initial
value, asparagine again appears in the plasma. This behavior has been confirmed
by experiments on leukemic patients.83 Even though the use of such reactors as
extracorporeal units can result in an effective decrease of asparagine or arginine
content in blood, its effect is limited to a few hours after the treatment. This ev-
idence suggests that more than one metabolite has to be removed from blood at
the same time to make the extracorporeal treatment effective.
Up to four enzymes involved in the metabolic pathways of purine bases-
allantoinase, allantoicase, uricase and catalase-have been immobilized together
by means of glutaraldehyde on the outer surface of cellulosic hollow fibers. Re-
actor performances are depicted in Figure 7.37, when uric acid is fed to the
multi-enzyme reactor.‘r
Enzyme stability and activity can be enhanced by performing enzyme im-

mobilization in the presence of inert proteins, such as albumin, or polyamines
in order to shift the maximum enzyme activity towards blood pH.
IOC
l Plasmatic L-Asn
nmoles/ml
50
l Piasmatic L-Glu
nmoleslml
400
0 Blood cells
L-Glu 200
nmoles/ml
Figure 7.36: L-Asparagine concentration vs. time in in vivo experiments per-

formed on rats. The extracorporeal device is a “tube-and-shell” membrane reac-
tor where asparaginase is bound to the outer surface of Cuprophan membranes.71
A Uric acid l Allantoin
CJAllantoic acid V Glyoxylic acid
Q-
P
‘t
I I 8 I I
I 2 3 4 5 6 7 a 2C Time h
Figure 7.37: Performance of a multi-enzyme membrane reactor when uric acid

is fed as a substrate. Allantoicase, allantoinase, uricase and catalase arecovalently
bound to the outer surface of a cellulosic membrane.71
Self-Cleaning Enzymatic Ultrafiltration Membranes

Thus far we have examined some of the possible applications of enzyme
membrane reactors, focusing attention on situations in which enzymatic conver-
sion is essential and the membranes act only as porous supports. However, im-
mobilized enzymes can also be used to improve the performance and effective-
ness of traditional ultrafiltration processes. Ultrafiltration of macromolecular
solutions, especially proteins, is strongly hindered by the so-called concentration
polarization and fouling phenomena (see Chapter 3 on ultrafiltration). The ir-
reversible deposition ” of a macromolecular gel layer on an active membrane sur-
face usually results in a sharp decay in permeate flux. Immobilization of pro-
teolytic enzymes onto the active side of the ultrafiltration membrane can pro-
vide a self-cleaning function. The hydrolytic cleavage of large proteins by pro-
teolytic enzyme is controlled by the amount of immobilized enzyme, the fluid
dynamic conditions under which the system is operated and by membrane re-
jection properties. As soon as macromolecules come in contact with the active
enzymatic layer, they are converted to low-molecular-weight products which
permeate the membrane.
Figure 7.3874 shows how transmembrane pressure can affect molecular
weight distribution in the permeate for the system dextran-dextranase when
enzymes are bound to polysulfone flat membranes by means of glutaraldehyde.
100
80
20
0
2 3 4 5 6
In MW
Figure 7.38: Sum curves of molecular weight distribution for different pres-
sures. Polysulfone membrane type 1. T = 40°C, pH = 6.0, dextran T70 1 wt%: (0)
0.5 bar; (0) 1 bar; (A) 3 bar; (17)4 bar.74
At low transmembrane pressure, macromolecules are nearly 100% converted to

products. As the pressure increases, macromolecular concentration in the per-
meate increases (see Chapter 3). By carefully choosing operating conditions, it
is possible to gain a net shift in molecular weight distribution curves (Figure
7.39)74
Even though enzymatic conversion is not too effective, it is possible to
prepare semipermeable membranes whose ultrafiltration yields are higher than
those of passive membranes.74r75 Ultrafiltration experiments of cheese whey
through cellulosic membranes to which papain was covalently bound, show that
flux decay curves of enzymatic membranes are even lesssensitive to pH changes.”
2 4 6
In MW
Figure 7.39: Sum curve of molecular weight distribution (C, %) in permeate and
reactor after 25 hr. Polysulfone membrane type 2, dextran T70 1.19 wt%, pH =
6.0. Blank symbols permeate; filled symbols reactor. (0) standardization solu-
tion, saccharose; (0) standardization solution, dextran T70; (0) 3 bar, (A) 5 bar.74
MEMBRANE FERMENTORS
Increasing interest is developing in continuous fermentation processeswhere

microporous membranes are used to separate the fermentation broth from the
product stream, thus retaining viable cells in the fermentor.
In 1970, Porter and Michael? first suggested the use of membranes to en-
hance fermentor productivity; traditional fermentors could in fact be coupled
to a membrane separation unit in a configuration called a “membrane fermen-
tar” where flux through the membrane was pressure driven. Since then a number
of different membrane configurations have been proposed. A typical apparatus

is shown in Figure 7.40 in which there is a vessel for the growing biomass, where
the pH and temperature are strictly controlled and nutrients are added, and a
UF unit, with membranes in hollow fiber or flat slab configuration, to withdraw
products from the flowing cell slurry.
LWd Control Proba

_Y
Ullratiltrolm7 Umt
F
Figure 7.40: Cell recycle fermentation apparatus.%
Viable cells can also be confined within the shell of a hollow fiber mem-
brane module8’ (Figure 7.41) or in a cell circuit separated with a dialysis mem-
brane from a dialysate circuit, where water or substrate solution is kept flowings6
(Figure 7.42). When kinetic models for the growth and fermentation of a spe-
cific kind of cell or microbe on the corresponding substrate medium are avail-
Product
Flowmeter
Hollow fibre module
Figure 7.41: Schematic of a hollow fiber fermentor (HFF).85

FERMENTOR DIALYSATE
CIRCUIT CIRCUIT
Figure 7.42: Schematic of dialysate-feed, immobilized cell system for dialysis

continuous fermentation.%
able, as is the case for Zymononas mobilis,8’ mathematical modeling of mem-

brane fermentors can be accomplished according to mass balances and the
equations reported for the corresponding enzyme reactor configurations ex-
amined in the previous sections. In Tables 7.1, 7.2 and 7.3, a summary of the
performance of different fermentor configurations is reported for three fermen-
tation processes, carried out by different microorganisms.
Ceil Recycle Fermentors

Cell recycle fermentors consist of two main units: a vessel where the bio-
mass is allowed to grow, and a membrane separation unit (as in Figure 7.40).
Vessels are usually designed to insure a uniform concentration of nutrients and
pH throughout the whole volume. Due to complete mixing, process control and
stability of the microbial slurry are not difficult to achieve.s8 After anaerobic
stabilization, when the biomass is well developed, the reactor biomass is pumped
to the UF unit where solid-liquid separation occurs. The sludge is flushed back
to the reactor. In most cases, the flow rate of nutrient feed is kept equal to the
permeate flow rate thus keeping a constant liquid level in the anaerobic reactor.
Table 7.1: Comparison of Performances of Different Fermentor

Configurations for Lactose to Ethanol Conversion by Kluyveromyces fragilis
HF means hollow fiber separation unit

Substrate Reac tar Sf Pr P Cell concentration Ref.
gfl g/lb g/l gfl
Lactose batch 50 3 22 3.8 89
Lactose CSTR+HF 50 25 15 3.8 a9
Lactose CSTR+HF 50 65 10 41 89
Lactose CSTR+HF I50 240 40 90 89
Lactose HFF 50 63 15-25 02 I35
Whey Lactose HFF 45 35 S-10 I17 85
Table 7.2: Comparison of Performances of Different Fermentors for Glucose

to Lactic Acid Transformation by Lactobacillus delbreuckii
HF means hollow fiber separation unit
Substrate Reactor Sf Pr P cell concentration Ref.
gfl g/l g/l g/l
Glucose batch 1-2 45 7-0 94
Glucose batch w/dialysis 2-3 35 II 94
Glucose CSTR 7 38 7-0 94
Glucose CSTR+HF 30 76 35 54 94
Glucose HF reactor 100 2 350 94
Glucose Ca. Alg. gel 3 46 67 94

beads
Table 7.3: Comparison of Performances of Different Fermentors for Glucose

Conversion to Ethanol by S. Cerevisiae
Substrate Reactor PK Ref.
gfl
Glucose dialysis 2.1-5.8 86
Glucose batch multiple I. 8-25 86
Glucose vacuum continuous 44 86
Glucose cant inuum with 53 86

cell recycle
Higher cell concentrations are usually achieved in cell recycle fermentors

than in the usual fermentor. Steady state mass balances on viable cells and on
the limiting growth substrates for a continuous fermentor (Figure 7.43) can be
written as: 11’
F X’ = Ll*x’ v (68)
and X' = D' Y (Sf - S)/ P* respectively.
Combining the two equations, the cell concentration within the vessel is:ll’
x’ = Y(Sf - S)
A steady state mass balance on a continuous cell recycle fermentor, over the ves-
sel alone is as follows (see Figure 7.40):
a’ F Xi - (1 + a') F Xi + V P* Xi = 0 (70)
on cells, and X’ = Y D’ (Sf - S)/ p* on limiting growth substrate to give a cell

concentration value in the vessel equal to:“’
Xi = Y (Sf - S)/(l + a’ - C a’) (71)
The cell concentration within the vessel of a cell recycle fermentor is then
greater by a factor of l/(1 t (Y’ - Co’).
The performance and capacity of an anaerobic reactor can be expressed in
terms of two parameters: the solid retention time (SRT), and the reactor volume
loading (VL).
The SRT is the average retention time of organisms in the reaction vessel.
For a suspended growth system, SRT is usually defined as the ratio of volatile
suspended solids (VSS) in the reactor to the VSS lost in the effluent or inten-
tionally wasted per day.88
The specific growth rate (SGR) is equal to the reciprocal of the SRT; it can
be expressed as a-linear function of the specific substrate utilization rate, KS, to
give:88
SGR = Ye KS-D
where Y = the organism yield coefficient: the massof cells formed/mass

nutrients
KS = the specific substrate utilization rate, mass/mass time
D = the organism decay coefficient, l/time.
The kinetics of substrate removal in the anaerobic reactor actually deter-

mines the SRT required for a given fermentation efficiency. In turn, the pres-
ence of the UF unit allows a fine control of SRT, making process control easier.
When high biomass concentrations are achieved, longer SRTs can be maintained
at lower reactor volumes.
L
f (literdr) X (&liter)
:.I I -
S (&liter)
V liquid volume (liters)
_ .
’ .
*. c. 3.‘.
s.
Figure 7.43: Schematic of a single stage chemostat.“’
The volumetric loading to a reactor, VL, is defined as:59
VL = 0 SflV = Sf/T
High biomass concentrations permit operation at higher volumetric loading

rates. Reactor design can be carried out by estimating the value of volumetric
loading necessary to achieve a given effluent quality. This design criterion is
commonly used for biological systems where the evaluation of biomass concen-
tration within the reactor is difficult. This is not the case with cell recycle fer-
mentors, so the VL or the SRT approach can also be used as design criterion.
Figure 7.44 shows the typical dependence at steady state of substrate and
product concentrations, and of productivity on permeate flow rate, i.e., dilu-
tion rate, for lactose fermentation to ethanol by Kluyveromices fragilis in a cell
recycle membrane fermentor.* As dilution rate increases fermentor productivity
increases, attains a maximum value and then decreases. Product and substrate
concentrations in the permeate, instead, steadily decrease and increase, respec-
tively, as dilution rate increases. A compromise generally has to be made be-
tween production rate and product concentration in the effluent. When the ab-
sence of substrate in the permeate is required, it obviously limits fermentor
productivity, as in the case of wastewater treatment in the dairy industry. On
the other hand, low substrate concentrations in the permeate keep recovery
-
L 240 ‘+
en
\
ci
E:
580
I
ilJ >
5
0 4 8
Dilution Rate 6, hr-l
Figure 7.44: Fermentation kinetics of a membrane recycle fermentor. Feed con-

centration = 150 g/Q lactose. Cell concentration = 90 g/P.se
costs of products from the effluent stream low. Therefore, the optimum operat-
ing conditions will be determined by the economics of the overall process.
Data in Table 7.1 show how the substrate concentration in the feed can
affect fermentor performance.
These data outline a dominant feature of the cell recycle membrane fermen-
tor: productivity is a monotonic function of viable cell concentration within the
fermentor. Cell recycle makes it possible to operate at higher microbial cell con-
centrations than in conventional batch fermentors, thus reducing the volume re-
quired for a given productivity and hence the capital costs.gOInvariably, the vis-
cosity of the cell slurry increases with cell concentration, so that concentration
polarization phenomena are usually significant. Complex systems such as rotat-
ing membrane fermentors have been proposed to overcome fouling and expected
concentration polarization problems.” In high rate fermentors, concentration
polarization is usually controlled by operating the reaction system at high re-
circulation rates though pumps have to be carefully chosen to avoid cell dam-
age.a6
When concentration polarization occurs, permeate fluxes become invariant
with the transmembrane pressure, and increases in the permeation rate can
be achieved only by enlarging the membrane surface area or by improved fluid
management. The design of the UF membrane unit in the polarized regime must
relate the flux to the optimal level of reactor VSS or total suspended solids
(TSS) according to Michaels’ gel theory, to give
J = KS log (C,/%
where Ks = the overall mass transfer coefficient,

cg = apparent solid concentration in the gel, and
x = anaerobic reactor VSS or TSS.
High recirculation and dilution rates help in maintaining low levels of in-
hibitory products. In ethanol fermentation processes, for instance, the removal
of these inhibitory products through the UF membrane can lead to significant
improvements in reactor performance. Both primary and secondary metabolites
freely pass through the membrane thus relieving both primary and secondary in-
hibition.” The use of a membrane able to remove ethanol selectively’* has been
proposed. In this way, primary inhibitors are released through the membrane,
but secondary products are allowed to accumulate; their concentration level
soon becomes toxic reducing cell viability and alcohol productivity.g3
Membrane Segregated Fermentors

Continuous fermentation processes can also be carried out in a hollow fiber
fermentor (HFF).s’ Cells are packed into the shell side of a hollow fiber module,
while substrate solution is fed to the core of the fibers (Figure 7.41). Since the
cell slurry is separated from the substrate solution by the membrane, HFFs can
be used for fermentation of liquid streams containing low molecular weight fer-
mentable substances. A careful choice of membrane molecular weight cut-off
can also assure that the cell environment is fully sterilized. Figure 7.45 shows
typical performance of an HFF for whey permeate fermentation to ethanol by
0 I * 1 I I1 I I I. * 15
2 4 6 8
Dilution Rate, d(hf’)
Figure 7.45: Fermentation kinetics of a hollow fiber fermentor in single pass

mode processing whey permeate (45 g/Q lactose concentration). Initial cell con-
centration = 117 g/L?.(0) lactose, (A) ethanol, (0) productivity.“’
Kluyveromices fragiis. The behavior of the fermentor is similar to that of a cell

recycle reactor with the same fermentation system. Higher productivity and
yields than with a batch fermentation are obtained (Tables 7.1, 7.2). The long
term cell stability in an HFF (Figure 7.46) is also better than in a batch fermen-
tor. However, even with HFF, reactor productivity is increased at the expense
of low substrate conversions, at least at low dilution rates.
In another membrane fermentor configuration, the cell slurry is separated
from substrate solution by a dialysis membrane86 (Figure 7.42). Two different
operational modes for this configuration have been proposed. In the first, sub-
strate is fed into a continuous fermentor circuit that is dialyzed against a contin-
uous dialysate circuit where water is kept flowing. In the second, substrate is fed
into the dialysate circuit and diffuses towards the batch fermentor circuit through
a dialysis membrane. Dialysis membrane fermentors provide a novel approach to
the immobilization of microbial cells and the relief of both primary and second-
ary inhibition, where it occurs. Substrate entering the fermentor is sterilized by
passage through the membrane and its consumption for cell growth is reduced to
metabolism maintenance. Moreover, cells are effectively immobilized. There are
some problems related to attaining steady state in terms of cell growth.% In ad-
dition, the solute exchange capacity of dialysis membranes limits both produc-
tivity and conversion (Table 7.2). Even though improvements can be obtained
by operating the fermentor according to the first suggested procedure, per-
formances are not yet comparable to those of other membrane fermentors.
0 80 160 240
Time,hr
Figure 7.46: Long term stability of a hollow fiber fermenter in single pass mode
processing whey permeate. Initial cell concentration is 100 g/P. Dilution rate (D’)
is changed from 2.7 to 1.3 Q/h after 80 hours of operation. (0) lactose, (A) eth-
anol, (a) productivity.”
Complex Fermentation Systems

In some cases, the substrate must be chemically or biochemically hydro-
lyzed to low-molecular-weight fermentable compounds prior to fermentation.
Two steps are then needed to carry out the overall transformation, namely hy-
drolysis and fermentation. Sometimes, raw substrate hydrolysis is conveniently
performed via an enzymatic route, as in the case of ethanol production from
cellulose or starch. This two-step transformation can be coupled into a single
which raw materials are continuously converted to valuable product.
60- o/
/
0
40-
/
20 -
/
0
/
OII
60
-e-o-e_4
60
t
40 _ 30 ;
-r-A-~-~ ;
_
2o :
20 l/l
10 :
0
o --Q-Q-Q-T- 0 m
O 0.25 0.5 0.75 1.0
DILUTION RATE (H-‘)
Figure 7.47: Continuous SSF fermentation using starch to produce liquified

starch stream of 130 g/Q total sugars with Z. mobilis ZM4 at pH 5.0 and 35°C.
(0) ethanol, (*) biomass, (0) reducing sugars.s*
Since the optimum temperature for the hydrolysis step is usually different
than that for fermentation, a “one-step transformation” may not be as efficient
as two steps.
Enzymes exhibit their highest activity at temperatures appreciably higher
than those required for microbial fermentations. When such differences are
negligible, cwzymatjr hydroJy+is and fermeniatjjon can shdtaneolldy OCCIJJ,
Provided that enzyme concentration is high enough to prevent enzymatic con-
version from being rate limiting, fermentation time and capital costs can be re-
duced.
What has been previously said about the advantages of continuous mem-
brane fermentors also applies to such complex systems. UF membranes can in
fact be used to recycle both enzymes and microbial cells thus increasing overall
system productivity.9’
Interesting applications of these systems can be found in ethanol produc-
tion from cellulose or starch. The overall process is called simultaneous saccharif-
ication and fermentation (SSF).98 In this process, two fermentors are connected
in series. The first one provides for the liquefaction of the slurry, which is then
pumped into the second fermentor, where simultaneous saccharification and
fermentation occur due to the presence of amyloglucosidase and 2. mobilis
cells.
Typical steady state performance of a continuous SSF membrane process is
reported in Figure 7.47. A comparison of continuous SSF to batch SSF shows
how performance can be improved with the use of membranes. Enzyme con-
sumption is strongly reduced when the reactor is operated in this way.
The continuous SSF membrane process also offers appreciable increases in
productivity, (though at the expense of reduced product concentration in the
effluent stream) as well as decreased energy requirements when compared to a
conventional two-step process.
REFERENCES
1. Lehninger, A.L., Biochimica, Bologna; Zanichelli (1976).

2. Laidler, K.J. and Bunting, P.S., The Chemical Kinetics of Enzyme Ac-
tion, Oxford; Qarendon Press (1973).
3. Trevan, Michael D., Immobilized Enzymes, New York; J. Wiley and Sons
(1981).
4. Chibata, I., Immobilized Enzymes: Halsted Press (1978).
5. Drioli, E. and Catapano, G., Chimica Oggi 7/B: 11-16 (1984).
6. Mosbach, K., Immobilized Enzymes. TIBS January l-3 (1980).
7. Wandrey, C., Enzyme Membrane Reactor Systems. Proceedings of the
International Conference on the Commercial Applications and Im-
plications of Biotechnology, pp. 577-588, Online Publications Ltd.,
Northwood, UK (1983).
8. Wisniewski, J., Winnicki, T. and Majewska, K., Biotechnoi. Bioeng. 25:
1441-1452 (1983).
9. Hong, J., Tsao, G.T. and Wankat, P.C., Biotechnol. Bioeng. 23: 1501-1516
(1981).
10. Wichmann, R., Wandrey, C., Buckmann, A.F. and Kula, M.R., Biotechnol.
Bioeng. 23: 2789-2802 (1981).
11. Hagerdal, B., Lopez-Leiva, M. and Matthiason, B., Desalination 35: 365-373
(1980).
12. Ohlson, I., Tragardh, G. and Hahn-Hagerdal, B., Biotechnol. Bioeng. 26:
647-653 (I 984).
13. Azhar, A. and Hamdy, M.K., Biotechnol. Bioeng. 23: 1297-1307 (1981).
14. Alfani, F., Cantarella, M. and Scardi, V., J. Membr. Sci. 16: 407-416 (1983).
15. Alfani, F., Cantarella, M., Erto, L. and Scardi, V., in: Energy from Biomass
(A. Strub, P. Chartier and G. Schleser, eds.), pp. 1000-1005, Applied
Science Publishers, London (1983).
16. Closset, G.P., Cobb, J.T. and Shah, Y.T., Biotechnol. Bioeng. 16: 345-360
(1974).
17. Tachauer, E., Cobb, J.T. and Shah, Y.T., Biotechnol. Bioeng. 16: 545-550
(1974).
18. Deeslie, W.D. and Cheryan, M., Biotechnol. Bioeng. 23: 2257-2270 (1981).
19. Cheryan, M. and Deeslie, W.D., JAOCS 60: 1112-I 115 (1983).
20. Leuchtenberger, W., Karrenbauer, M. and Plocker, U., Scale up of an En-
zyme Membrane Reactor Process for the Manufacture of L-Enantiomeric
Compoungs, presented at the Europe-Japan Congress on Membranes and
Membrane Technology, Stresa, Italy (1984).
21. Wandrey, C., Wichmann, R. and Jandel, A.S., in: Enzyme Engineering (I.
Chibata, S. Fukui and L.B. Wingard, Jr., eds.), Vol. 6, pp 61-68, Plenum
Press, New York and London (1982).
22. Wandrey, C., Wichmann, R., Leuchtenberger, W., Kula, M.R. and Buckmann,
A., U.S. Patent 4,304,858 (1982).
23. Kim, I.H, and Chang, H.N.,AIChEJ. 29: 910-914 (1983).
24. Kim, I.H. and Chang, H.N., AlChE J. 29: 645-651 (1983).
25. Urabe, I., Katayama, M. and Okada, H., in: Enzyme Engineering (I. Chibata,
S. Fukui and L.B. Wingard, Jr., eds.), Vol. 6, pp. 239-240, Plenum Press,
New York and London (1982).
26. Deeslie, W.D. and Cheryan, M., Biotechnol. Bioeng. 24: 69-82 (1982).
27. Drioli, E., Gianfreda, L., Palescandolo, R. and Scardi, V., Biotechnol. Bioeng.
17: 1365 (1975).
28. Drioli, E. and Scardi, V., J. Membr. Sci. 1: 237 (1976).
29. Capobianco, G., Drioli, E. and Ragosta, G., J. of Solid-Phase Biochem. 2:
315-328 (1977).
30. Greco, Guido, Jr., Alfani, F., lorio, G., Cantarella, M., Formisano, A.,
Gianfreda, L., Palescandolo, R. and Scardi, V., Biotechnol. Bioeng.
21: 1421-1438 (1979).
31. Rossi, M., Nucci, R., Raia, C.A., Molinari, R. and Drioli, E., J. Mol. Catal.
4: 233-236 (1978).
32. Cantarella, M., Gianfreda, L., Palescandolo, R., Scardi, V., lorio, G., Alfani,
F. and Greco, Guido, Jr., Quaderni dell’lngegnere Chimico Italian0 61:
115-123 (1979).
33. Cantarella, M., Gianfreda, L., Palescandolo, R., Scardi, V., Greco, Guido,
Jr., Alfani, F. and lorio, G., J. of Solid-Phase Biochem. 2: 163-174
(1977).
34. lorio, G., Catapano, G., Drioli, E., Rossi, M. and Rella, R., J. Membr.
Sci. 22: 317-324 (1985).
35. Alfani, F., lorio, G., Greco, Guido, Jr., Cantarella, M., Remy, M.H. and
Scardi, V., Chem. Eng. Sci. 34: 1213-1219 (1979).
36. lorio, G., Catapano, G., Drioli, E., Rossi, M. and Rella, R., Ann. N. Y.
Acad. Sci. 434: 123-l 26 (1984).
37. Alfani, F., Albanesi, D., Cantarella, M. and Scardi, V., Enzyme Microb.
Technol. 4: 181-184 (1982).
38. Virkar, P.D., Narendranathan, T.J., Hoare, H. and Dunnill, P., Biotechnol.
Bioeng. 23: 425-429 (1981).
39. Alfani, F., Albanesi, D., Cantarella, M. and Scardi, V., Biotechnol. Left.
3: 309-314 (1981).
40. Mansi, F. and Curcio, F., Processi di Denaturazione e Stabilizzazione in
Reattori Enzimatici a Membrana Ultrafiltrante. Laurea Degree Thesis,
Dept. of Chem. Eng., Fat. of Engineering, Univ. of Naples, Italy (1980).
41. Greco, Guido, Jr., Albanesi, D., Cantarella, M. and Scardi, V., Biotechnol.
Bioeng. 22: 215-219 (1980).
42. Engasser, J.M., Caaron, J. and Marc, A., Chem. Eng. Sci. 35: 99-105 (1980).
43. Engasser, J.M., Biochim. Biophys. Acta 526: 301-310 (1978).
44. Klei, H.E., Sundstrom, D.W., Coughlin, R.W. and Ziolkowski, K., Biotech-
no/. and Bioeng. Symp. 11: 593-601 (I 981).
45. Glomon, C., Germain, P., Miclo, A. and Engasser, J.M., in: Enzyme En-
gineering (I. Chibata, S. Fukui and L. B. Wingard, Jr., eds.), Vol. 6,
pp. 125-126, Plenum Press, New York and London (1982).
46. Rony, P.R., Biotechnol. Bioeng. 13: 431-447 (1971).
47. Waterland, L.R., Michaels, A.S. and Robertson, CR., AIChE J. 20: 50-59
(1974).
48. Kitano, H., Yoshijima, S. and Ise, N., Biotechnol. Bioeng. 22: 2643-2653
(1980).
49. Kitano, H., Yoshijima, S., Hotogi, S. and Ise, N., Biotechnol. Bioeng. 22:
2633-2642 (1980).
50. Fink, D.I. and Rodwell, V.W., Biotechnol. Bioeng. 17: 1029-1050 (1975).
51. Davis, J.C., Biotechnol. Bioeng. 16: 1113-l 122 (1974).
52. Rony, P.R., J. Am. Chem. Sot. 94: 8247-8248 (1972).
53. Lewis, W. and Middleman, S., AlChE J. 20: 1012-1014 (1974).
54. Korus, R.A. and Olson, A.C., Biotechnol. Bioeng. 19: l-8 (1977).
55. Korus, R.A. and Olson, A.C., J. Food Sci. 42: 258-260 (1977).
56. Miyawaki, O., Nakamura, K. and Yano, T., J. Chem. Eng. Jpn. 15: 142-147
(1982).
57. Sparks, R.E., Mason, N.S., Finley, T.C. and Scharp, D.W., Trans. Am. Sot.
Artif. Intern. Organs 28: 229-231 (1982).
58. Reach, G., Jaffrin, M.Y. and Deajeux, U.F., A U-Shaped Flat Membrane
Bio-Artificial Pancreas. Application of the Ultrafiltration-Based Model
to Its Design. Proceedings of the European Sot. of Artif. Intern. Organs,
pp. 49-52 (1983).
59. Reach, G., Jaffrin, M.Y. and Assan, R., Life Support Systems, 1, Supp.:
73-75 (1982).
60. Kastl, P.R., Baricos, W.H., Chambers, R.P. and Cohen, W., private communi-
cation (1978).
61. Drioli, E., lorio, G., De Rosa, M., Gambacorta, A. and Nicolaus, B., J.
Membr. Sci. 11: 365-370 (1982).
62. Breslau, B.R. and Kelcullen, B.M., Hollow Fiber Enzymatic Reactors. An
Engineering Approach presented at the Third International Conference
on Enzyme Engineering, Portland, Oregon (1975).
63. Colton, C.K., Solomon, B.A., Galletti, P.M., Richardson, P.D., Takahashi,
C., Naber, S.P. and Chick, W.L., in: Ultrafiltration Membrane and Appli-
cations (A.R. Cooper, ed.), 541-555, Plenum, New York, London (1980).
64. Chambers, R.P., Cohen, W. and Baricos, W.H., Methods in Enzymology 44:
291-317 (1976).
65. Craveri, R., Fabiani, C., Leonardi, M. and Pizzichini, M., Performance of
Whole Cells possessing Cellobiase Activity Immobilized into Hollow
Fiber Membrane Reactors presented at the Europe-Japan Congress on
Membranes and Membrane Processes, Stresa, Italy (1984 ).
66. Orsetti, A., Bouhaddioui, N., Finley, T.C., Crespy, S. and Perez, R., C.R.
Sot. Biol. 175: 189-201 (1981).
67. Drioli, E., lorio, G., Catapano, G., De Rosa, M. and Gambacorta, A., J.
Membr. Sci. 27: 253-261 (1986).
68. Drioli, E., lorio, G., Molinari, R., De Rosa, M., Gambacorta, A., Nicolaus,
B. and Esposito, E., Biotechnol. Bioeng. 23: 221-223 (1981).
69 Drioli, E., lorio, G., Santoro, R., De Rosa, M., Gambacorta, A. and Nicolaus,
B., J. Mol. Catal. 14: 247-251 (1982).
70. Callegaro, L. and Rossi, V., in: Hollow Fibers and Capillary Membranes in
New Separation Processes (C.N.R. of Italy ed.), pp. 44-51 (1981).
71. Mazzola, G. and Vecchio, G., in: Hollow Fibers and Caoillary Membranes
in New Separation Processes (C.N.R. of Italy ed.), pp. 52-58 (1981).
72. Mazid, M.A. and Laidler, K.J., Biotechnol. Bioeng. 24: 2087-2097 (1982).
73. Bunting, P.S. and Laidler, K.J., Biotechnol. Bioeng. 16: 119-134 (1974).
74. Staude, E., Jorisch, W. and Ansorge, W., J. Membr. Sci. 11: 289-296
(1982).
75. Howell, J.A., Knapp, J.S. and Velicangil, 0.. in: Enzyme Engineering (G.B.
Broun, G. Manecke and L.B. Wingard, Jr., eds.), Vol. 4, pp. 267-271,
Plenum Press, New York and London (1978).
76. Kobayashi, T. and Laidler, K.J., Biotechnol. Bioeng. 16: 77-97 (1974).
77. Vieth, W.R., Venkatasubramanian, K., Constantinides, A. and Davidson,
B., in: Applied Biochemistry and Bioengineering (L.B. Wingard, E.
Katchalsky-Katzir and L. Goldstein, eds.), Vol. 1, Academic Press, New
York (1976).
78. Chang, T.M.S., in: Biomedical Applications of Immobilized Enzymes and
Proteins (T.M.S. Chang, ed.), p. 105, Plenum Press, New York (1977).
79. Rossi, V., Int. J. Art. Organs 4: 102-I 07 (1981).
80. Rossi, V., ISA0 Congress, Paris, France (1981).
81. Horvath, C., Shendalman, L.H. and Light, R.T., Chem. Eng. Sci. 28: 375-
388 (1973).
82. Adu-Amankwa, B. and Constantinides, A., Biotechnol. Bioeng. 26: 156-
166 (1984).
83. Giordano, C., Esposito, R., Mazzola, G., Vecchio, G., Pluvio, G., Cirillo,
D., Capasso, G.B. and Buonanno, G., lnt. J. Artif. Organs 4: 62-67 (1981).
84. Porter, M.C. and Michaels, AS., Chem. Tech. 2: 56-61 (1970).
85. Mehaia, M.A. and Cheryan, M., Enzyme Microb. Technol. 6: 117-120
(1984).
86. Kyung, K.H. and Gerhardt, P., Biotechnol. Bioeng. 26: 252-256 (1984).
87. Slotnicki, M.L., Warr, R.G., Goodman, A.E., Lee, K.J. and Rogers, P.L.,
in: Biotechnology (C.F. Phelps and P.H. Clarke, eds.), Vol. 48, pp.
53-86, Biochem. Sot. Symp., Great Britain (1983).
88. Li, A. and Sutton, P.M., Dorr-Oliver’s Membrane Anaerobic Reactor Sys-
tem. internal Report, Dorr-Oliver (1982).
89. Cheryan, M. and Mehaia, M.A., Biotechnol. Lett. 5: 519-524 (1983).
90. Maiorella, B., Wilke, Ch.R. and Blanch, H.W., A& Biochem. Eng. 20:
43-92 (1981).
91. Margaritis, A. and Wilke, Ch.R., Biotechnol. Bioeng. 20: 727-753 (1978).
92. Gregor, H. and Jeffries, T.,Ann. N. Y. Acad. Sci. 326: 273 (1979).
93. Maiorella, B., Blanch, H.W. and Wilke, C.R., Biotechnol. Bioeng. 25:
103-121 (1983).
94. Vick Roy, T.B., Mandel, D.K., Dea, D.K., Blanch, H.W. and Wilke, CR.,
Biotechnol. Lett. 5: 665-670 (1983).
95. Janssens, J.H., Bernard, A. and Bailey, R.B., Biotechnol. Bioeng. 26:
l-5 (1984).
96. Dostalek, M. and Haggstrom, M., Biotechnol. Bioeng. 24: 2077-2086
(1982).
97. Charley, R.C., Fein, J.E., Lavers, B.H., Lawford, H.G. and Lawford, G.R.,
Biotechnol. Lett. 5: 169-174 (1983).
98. Lee, J.H., Pagan, R.J. and Rogers, P.L., Biotechnol. Bioeng. 25: 659-
669 (1983).
99. Levenspiel, O., Chemical Reaction Engineering, New York, London: J.
Wiley & Sons (1972).
100. Michaels, A.S., Chem. Eng. Progr. 64: 31-43 (1968).
101. Blatt, W.F., Dravid, A., Michaels, A.S. and Nelsen, L., in: Membrane
Science and Technology (J.E. Flinn, ed.), pp. 40-97, Plenum Press,
New York (1970).
102. Mathiason, E. and Sivik, B., Desalination 35: 59-103 (1980).
103. Lowe, CR. and Mosbach, K., Eur. J. Biochem. 49: 51 I-520 (1974).
104. Zappelli, P., Rossodivita, A. and Re, L., Eur. J. Biochem. 54: 475-482
(1975).
105. Adamski-Medda, D., Nguyen, Q.T. and Declacherie, E., J. Membr. Sci. 9:
337-342 (1981).
106. Greco, Guido, Jr., Marrucci, G. and Gianfreda, L., in: Enzyme Engineering
(I. Chibata, S. Fukui and L. B. Wingard, Jr., eds.), Vol. 6, pp. 229-232,
Plenum Press, New York and London (1982).
107. Porter, M.C., lnd. Eng. Chem. Prod. Res. Dev. 11: 234-248 (1972).
108. lorio, G., Drioli, E., Memoli, B., Andreucci, V.E., Salvatore, M. and Alfano,
B., J. Membr. Sci. 18: 297-311 (1984).
109. Pitcher, W.H., Jr., and Havewala, N.B., in: Enzymology (H.H. Weetcell,
ed.), pp. l-93 (1974).
110. Davis, E.J., Cooney, D.O. and Chang, R., Chem. Eng. J. 7: 213-225 (1974).
111. Wykes, J.R., Dunnill, P., Lilly, M.D., Biochim. Biophys. Acta 286: 260-
268 (1972).
112. Messing, R.A., in: Immobilized Enzymes for Industrial Reactors (R.A.
Messing, ed.), pp. 79-99, Academic Press, New York (1975).
113. Weethali, H.H., in: Immobilized Enzymes for Industrial Reactors (R.A.
Messing, ed.), pp. 99-123, Academic Press, New York (1975).
114. Grubhofer, S. and Schleith, L., Hoppe-Seyler’s Z. Physiol. Chem. 297:
108 (1954).
115. Mascini, M., lannello, M. and Palleschi, G., Anal. Chim. Acta 146: 135-
148 (1983).
116. Karube, I. and Suzuki, S., in: Enzyme Engineering (I. Chibata, S. Fukui
and L.B. Wingard, Jr., eds.), Vol. 6, pp. 417-418, Plenum Press, New
York and London (1982).
117. Wang, I.C.D., Cooney, C.L., Demain, A.L., Dunnill, P., Humphrey, A.E.
and Lilly, M.D., Fermentation and Enzyme Technology, New York:
J. Wiley & Sons (1979).
8
Electrodialysis
Thomas A. Davis
INTRODUCTION
Electrodialysis (ED) is an electromembrane process in which ions are trans-

ported through ion-exchange membranes from one solution to another under the
influence of an electrical potential. The nature of the membranes and the driv-
ing force distinguish ED from the pressure driven membrane processes such as
gas permeation, reverse osmosis, and filtration. The electrical charges of ions al-
low them to be driven through solutions and water-swollen membranes when a
voltage is applied across these media. Since pressures are usually balanced across
the membranes, the requirements for strength and support of the membranes
and containment vessels are less demanding in electromembrane processes than
in pressure driven membrane processes. Instead, design emphasis is directed to-
ward maintaining uniform distribution of solution flow and minimizing elec-
trical resistance and current leakage.
Some important concepts of electromembrane processes are presented in
Figure 8.1. Ionic transport is illustrated in Figure 8.la with no membranes pres-
ent. When NaCl is dissolved in water, it dissociates into Na+ cations and Cl-
anions. Application of an electric potential to the electrodes causes the ions to
move through the solution at velocities proportional to the strength of the elec-
tric field. Average ionic velocities in the bulk solution are surprisingly slow, usu-
ally less than 1 cm/min, but the ions do not have to travel very far to reach the
membranes in the thin solution compartments of ED stacks. The combined
motion of Na’ to the right and Cl- to the left carries the total electric current
flow through the bulk solution (unless other ions are present). Electrode reac-
tions, which will be discussed in detail, transfer the current from the solution
to the anode and cathode. Clz gas is generated at the anode while Hz gas and
OH- ions are generated at the cathode.
One could produce C12, Hz and NaOH from the electrolytic cell in Figure
8.la. but the products would be impure. In conventional electrolytic cells, a
porous diaphragm between the electrodes prevents bulk mixing of the products
482
Eiectrodialysis 483
of the electrode reactions. Modern chloralkali cells utilize a cation-exchange

membrane between the electrodes as illustrated in Figure 8.lb. This innova-
tion has substantially improved the purity of the caustic produced. The cation-
exchange membrane is permeable to Na’ ion but relatively impermeable to
Cl- and OH- ions. Therefore, nearly all of the current is utilized to produce CIZ
and NaOH. This membrane electrolysis cell is expected to dominate chloralkali
plants in the future.
When a pair of anion and cation-exchange membranes is placed between the
electrodes as shown in Figure 8.1~. the solution between the membranes be-
comes depleted of NaCI. This depletion occurs because the membranes are se-
lectively permeable to ions of a specific charge. The anion-exchange membrane
allows Cl- ions to be carried out of the center compartment by the electric po-
tential, but it does not allow Na’ ions to enter. Similarly the cation-exchange
membrane offers a low resistance to Na’ ions and a high resistance to Cl- ions.
If the polarity (charge) of the electrodes were reversed or if the positions of
the two membranes in Figure 8.1~ were switched, the NaCl content of the cen-
ter compartment would increase. This effect is accomplished in an ED stack by
an array of alternating anion- and cation-exchange membranes as shown in Fig-
ure 8.ld. The solutions between the membranes are alternately enriched in or
depleted of NaCl when the electrodes are energized. The enriched and depleted
solutions are withdrawn from their respective compartments to achieve useful
changes in the electrolyte content of solutions without substantially affecting
the content of nonelectrolytes. It is this selectivity for electrolytes that often
makes ED the process of choice for certain separations, e.g., desalting of protein
solutions or whey.
Ion-Exchange Membranes
The ion-exchange membranes used in electrodialysis are essentially sheets
of ion-exchange resins. Figure 8.2 illustrates the structure of a cation-exchange
membrane which has negatively charged groups (e.g., -SO;) chemically attached
to the polymer chains (e.g., polystyrene). Ions with a charge opposite to the
fixed charge (counter ions) are freely exchanged at these sites. The concentra-
tion of counter ions (e.g., Na’) is relatively high; therefore, counter ions carry
most of the electric current through the membrane. The fixed charges attached
to the polymer chains repel ions of the same charge (co-ions), in this case the
anions. Since their concentration is very low, anions carry only a small fraction
of the electric current through a cation-exchange membrane. Attachment of
positive fixed charges (e.g., -NRs+) to the polymer chains forms anion-exchange
membranes, which are selectively permeable to negative ions, because the fixed
-NRs’ groups repel positive ions. This exclusion, as a result of electrostatic re-
pulsion, is called Donnan exclusion.
Since ion-exchange polymers (e.g., styrenesulfonic acid) are water soluble,
crosslinking is needed to prevent dissolution of ion-exchange membranes. Di-
vinylbenzene is used to crosslink styrene. The degree of crosslinking and the
fixed-charge density affect the membrane’s properties in opposite ways. Higher
crosslinking improves selectivity and membrane stability by reducing swelling,
but it increases electrical resistance. High charge density reduces resistance
and increases selectivity, but it promotes swelling and thus necessitates higher
cl_-
I
NaCl
a) ELECTROLYTIC CONDUCTION b) ADDITION OF CATION-

WITH INEFFICIENT EXCHANGE MEMBRANE
ELECTROLYSIS ALLOWS EFFICIENT
ELECTROLYSIS OF NaCl
FOR Cl AND NaOH
PROD&ION
CONCENTRATED NaCl
=qqf%qf%
A C
DLUTE
-1
N&l
+
+ Cl- Na+ c-
1
tt N&l t t t t
t
NaCI FEED
d) ELECTRODIALYSIS:
MULTIPLE MEMBRANE
PAIRS BETWEEN A SINGLE
SALT OEPLET ION OCCURS PAIR OF ELECTRODES
BETWEEN A PAIR OF EFFICIENTLY USE CURRENT
ANION- AND CATION- TO PRODUCE DILUTE AND
EXCHANGE MEMBRANES CONCENTRATED NaCl
SOLUTIONS
Figure 8.1: Membranes affect ionic transport. (a) Electrolytic conduction with
inefficient electrolysis. (b) Addition of cation-exchange membrane allows
efficient electrolysis of NaCl for Cl* and NaOH production. (c) Salt depletion
occurs between a pair of anion- and cation-exchange membranes. (dj Electro-
dialysis: multiple membrane pairs between a single pair of electrodes efficiently
use current to produce dilute and concentrated NaCl solutions.
crosslinking. A compromise among selectivity, electrical resistance, and dimen-

sional stability is achieved by proper adjustment of crosslinking and fixed-charge
densities.
Membrane Types
There are two general types of commercially available ion-exchange mem-
branes: heterogeneous and homogeneous. Both types usually contain a rein-
forcing fabric to increase tensile strength and improve dimensional stability.
Heterogeneous membranes have two distinct polymer phases. They are rather
Electrodialysis 485
0 FIXED NEGATIVELY CHARGED EXCHANGE SITE, I.&. SO;

(B MOBILE POSITIVELY CHARGED EXCHANGEABLE CATION; I.E., Na+
=POLYSTYRENE CHAIN
=X~IVINYLBEN~ENE CROSSLINK
Figure 8.2: Diagram of cation-exchange membrane.
easily made by grinding up ion-exchange resins and dispersing the particles in a

film-forming polymer. Most commercially available membranes are of the ho-
mogeneous type with a continuous polymer phase containing ionic groups at-
tached to the polymer chains. The most common methods of membrane prep-
aration are listed below.
Heterogeneous. Ion-exchange resin particles dispersed in polymer film.
l Calender or press mixture of resin and polymer.
0 Cast film from dispersion of resin in polymer solution.
l Disperse resin in prepolymer, cast film, and polymerize.

Homogeneous. ionic groups attached to polymer.
l Polymerize ionic monomers.

l Crosslin k polyelectrolyte.
l Graft onto preformed films.
0 Cast solution of polyelectrolyte in film-forming polymer.
l Imbibe graftable monomers into film, polymerize, and graft.
Membrane Properties
The physical properties of some commercially available ion-exchange mem-
branes are listed in Table 8.1. The information on membrane properties was that
supplied by manufacturers unless otherwise indicated. The diversity of mem-
brane thickness is noteworthy. The membranes made by lonics, Inc. are thick
and rigid, because they must span rather wide spaces between supports in the
solution compartments without deflection. The thick lonics membranes have
relatively high resistance, but these values are still low compared to the resis-
tance of the brackish water they are designed to treat. RAI developed very thin
membranes for battery separators and used the same radiation grafting tech-
niques to make thin ED membranes with low resistance.
Table 8.1: Commercial Ion-Exchange Membranes
THICKNESS COUNTERIONb
-TYPE um TRANSPORT, %
&SW, CHEMICAL
7owo. ,*P*N
C 240 2.1 95
A 210 2.5 73
ASAH, GLASS
7OKVO. ,*P*N
CM” C 135 2.7 91

AM” A 135 2.7 93
IONAC CHEMICAL
BIRMINGHAM.N.J., USA
MC-3470 C 3b3 9.6 e3=

MA-Y7SR A 368 10.5 SC
IONICS. INC.
WATERTOWN, MA., USA
CR-61 CZL-3% .C 506 10 90.3

AR 103 OZL-366 A 506 12.5 66.6
AR 20) ULL-366 A 506 9 95.0
RAI RESEARCH CORP.

HAUPPAGE. N.Y..USA
R-4010 C 114 66
R-4035 A 69 66
TOKUVAHA SODA
TOKYO, JAPAN
C66-51 C 15s 9&c

*F-&T A 175 9s=
a. RESISTANCE MEASURED IN I.ON KC,
b. TRANSPORT NUMBER MEASURED BETWEEN 0.5 ANO ,.ON KC,
C. DATA FROM REFEREUCC 2. ALL OTHER DATA SUPPLIED BY MANUFACTURERS.
Electrodialysis 487
The transport number of the counter ion provides an indication of the

permselectivity of the membrane. However, the conditions of measurement must
be specified, because the counter-ion transport number drops as the concentra-
tion of the external solution increases. Techniques for measuring the transport
number are described by Helfferich.’
The membrane properties reported in Table 8.1 are for new membranes. A
long-term study by Kneifel and Hattenbach’ revealed deterioration in some of
these properties after prolonged exposure to 0.1 N solutions of NaCI, NaOH, and
HN03. NaOH tended to be most destructive.
Water Transport
The permeability of ion exchange membranes to water is seldom tabulated
as a membrane property, because it varies so much with types and concentra-
tions of ions. Water transport is not often a significant factor in ED of dilute
solutions. However, water transport determines the upper concentration limit
of the enriching stream when ED is used to concentrate solutions. Ions that are
transported through the membranes by an electric potential drag along about
five water molecules on the average.3 When solutions are dilute and membranes
are loosely crosslinked, substantially more water transport occurs. Membranes
that are sufficiently crosslinked to restrict water transport in dilute solutions are
referred to as “tight” membranes.
ED STACKS
Design Considerations
An ED stack consists of alternating anion- and cation-exchange membranes
with solution compartments between them. The solution compartments are
bounded by perimeter gaskets that are pressed tightly between the membranes
to confine the solutions within the compartments. The compartments usually
contain spacers that keep the membranes separated by a constant distance. The
spacers also aid in distributing the solution velocity evenly throughout each com-
partment. (Poor flow distribution within and among the compartments severely
limits the performance of an ED stack.) Each solution compartment has a means
for introduction and removal of solutions. All of these features are combined
into a plastic device called a gasket-spacer. Most ED stacks are designed to use
identical gasket-spacers for enriching and depleting compartments.
An exploded view showing the components of an electrodialysis stack is
presented in Figure 8.3. The end plate is typically made of rigid plastic. It con-
tains an electrode and the connections for the solution that rinses the electrode.
It also contains connections for solution streams into and out of the enriching
and depleting compartments. Holes in the inside face of the end plate are aligned
with holes in the membranes and gasket-spacers that form manifolds for solu-
tion flow through the ED stack. Large stacks have multiple manifold holes on
both ends of the gasket-spacers to provide uniform flow over the width of the
compartment. The two membranes and two gasket-spacers illustrated in Figure
8.3 constitutes one cell pair. Commercial ED stacks typically contain more than
100 cell pairs between a single pair of electrodes.
-
-
1
ANION- w.N”
MORE
EXC”*NGE MEUBRANES,
SPKEI Fiu.MES,
MEMBRANE AND ANOTHER
EN0 FRAME
_
E
-
- -
Figure 8.3: Main components of an electrodialysis stack.
The ED stack in Figure 8.3 represents the sheet-flow type. The sheet-flow
design is used by most ED manufacturers (lonics, Inc. is the major exception).
The spacing between membranes of sheet-flow stacks is made as thin as practical
to reduce electrical resistance and minimize the thickness of hydraulic boundary
layers at the membrane surfaces. The thinness is ultimately limited by hydraulic
resistance and by problems associated with getting solutions into and out of the
solution compartments between the membranes. To maintain constant spacing
between the membranes it is necessary to support the membranes at close inter-
vals, but the support material must not cover a large fraction of the membrane
surface or cause stagnation in the solutions near the membrane surface. The
Vexa@ non-woven polyolefin netting developed by DuPont has gained wide
acceptance as a spacer that meets these criteria. The flow of solution in the plane
of Vexar netting is illustrated in Figure 8.4.
The stack design developed by lonics, Inc. is quite different from the sheet-
flow stacks previously described. The lonics design utilizes the tortuous-path
spacer illustrated in Figure 8.5 to direct solutions over a relatively long, open
channel in the solution compartment. Superimposed on this channel is a spacer
that supports the membrane and promotes turbulence in the solution compart-
ment. These supports are more widely spaced than those of a sheet-flow design,
so a more rigid membrane is required to bridge the space without severe deflec-
tion. Therefore, the membranes made by lonics are substantially thicker than
those produced by most other manufacturers. The thick membranes and spacers
result in higher electrical resistances than those found in sheet-flow stacks, but
the openness of the solution channels should make them less prone to fouling
by suspended solids.
Yet another stack design utilizes unit cells. Unit-cell stacks were specifically
developed for concentrating solutions. Each concentrating cell consists of one
cation-exchange membrane and one anion-exchange membrane sealed at the
edges to form an envelope with a spacer screen inside. The envelopes also have
Electrodialysis 489
Figure 8.4: Representation of solution flow through Vexar spacers.
Figure 8.5: Diagram of a tortuous-path spacer for an electrodialysis stack.

spacer screens between them to insure uniform solution flow across the mem-
branes. The entire assembly of alternating membrane envelopes and spacer
screens is held between a set of electrodes. The electric current carries ions from
the external solutions through the membranes to the inside of the envelopes
where they are trapped. Only osmotically and electro-osmotically transferred
water flows through the membranes. Thus, unit cells achieve the maximum de-
gree of concentration that is possible with ED. The concentrated solutions inside
of the concentrate cells flow through small tubes that lead from inside the con-
centrate cells to a plenum chamber arranged outside the stack.
STAGING OF ED STACKS
ED differs from other desalting processes in the degree of desalting achieved

in a single stage. In evaporation, reverse osmosis, and ion exchange the necessary
degree of desalting can often be achieved with one pass through the device. In
ED, the degree of desalting is usually limited to about 50% per pass, and some
type of staging is needed for further desalting. In large installations such as mu-
nicipal water supplies this staging is achieved by passing the water through a
series of stacks. For example, water with 2,000 ppm TDS (total dissolved sol-
ids) could be reduced to 500 ppm TDS by two stacks that each achieve 50%
desalting.
For small installations where the throughput does not justify multiple
stacks, there are other alternatives including internal staging, feed and bleed, and
batch recirculation. Internal staging allows one to utilize one pair of electrodes
to achieve multi-stage desalting. For the example cited above, one might use a
stack with 150 cell pairs to treat water with 2,000 ppm TDS. The first stage
would utilize 100 cell pairs to reduce the salinity to 1,000 ppm. Then a sec-
ond stage with 50 cell pairs would reduce the salinity to 500 ppm. Each cell
pair in the second stage would remove the same quantity of salt as a cell pair in
the first stage, but with half the TDS reduction because the water throughput
per cell would be doubled.
With feed and bleed or batch recycle systems some or all of the water that
has already been processed by the stack is mixed with feed and returned via a
recirculation pump. Recycle is inherently less efficient than once-through flow
in both stack utilization and in energy consumptions since the same solution
must be pumped and desalted repeatedly and then remixed with a more con-
centrated solution. However, the increased flexibility in process control makes
recycle systems attractive for small-scale operations where stacks are over-sized
to handle varying loads. Feed and bleed is useful for operations where the en-
riching stream needs to be as concentrated as possible. Ten-fold enrichment or
90% recovery of some feed waters as diluate can be achieved by concentrate re-
circulation if solubility limits of the dissolved substances are not exceeded in the
concentrate stream.
APPLICATIONS OF ED
Outside of Japan, the largest application of ED has been in the desalination

Electrodialysis 491
of brackish water. Most of the early research and development effort had the ob-
jective of improving marginal waters to make them potable. During the 1950’s,
the Netherlands National Research Organization, TNO, the South African Coun-
cil for Scientific and Industrial Research, and the Office of Saline Water in the
U.S.A. funded programs to improve membranes, equipment and processes to
produce potable water more economically. Since ED treatment of brackish wa-
ter is well known, that subject will not be stressed in this chapter.
A potential growth area for ED is in the rough desalting of water that will
be subjected to subsequent purification for use as boiler feed or rinse water in
the electronics industry. Ion exchange has traditionally been used for preparing
waters with low salinity, but the cost of regenerants and the magnitude of the
waste disposal problem are proportional to the salinity of the feedwater. The
bulk of the dissolved solids can be removed more economically by ED or RO.
These two processes are competitive in cost, and both offer the advantage that
they do not contribute additional water pollutants.
ED might appear to be at a competitive disadvantage with RO for this ap-
plication, because RO removes particulates and substantial amounts of organic
solutes from the water. Accumulation of these contaminants on membrane sur-
faces can cause premature failure of RO modules. Therefore, it is common prac-
tice to remove organics and particulates before the rough desalting. With such
pretreatment, ED stacks operate trouble-free for years with little maintenance.
The system shown schematically in Figure 8.6 has been in service at Bell Lab-
oratories since 1975 removing salts, principally CaS04, from water for use in
rinsing integrated circuits.4
ED desalting of seawater has also been demonstrated on a commercial scale.
The power consumption has been held to tolerable levels (28 watt-hours/gal) by
the use of ion-exchange membranes with low resistance.’ Still the greatest use
of ED treatment of seawater has been for the recovery of NaCI. Japan has no
natural deposits of NaCl and had formerly obtained the salt by evaporation of
seawater. The use of ED to produce a concentrated brine up to 20% TDS has
substantially reduced energy costs (less than 200 kwh per ton of salt). Moreover,
the use of membranes that are selective for univalent ions has improved salt
purity to 97% and reduced precipitation problems. The approach has been so
successful that essentially all the table salt in Japan is prepared from ED en-
riched brines.6
Food processing provides many potential applications for ED because of
its ability to separate electrolytes from non-electrolytes. Its greatest use has been
in whey deashing. Whey is the waste product from cheese making, and it con-
tains useful quantities of proteins, lactose, and lactic acid. However, the high
mineral content makes it unacceptable for human consumption and of marginal
value as animal feed. Deashing by ED upgrades the whey so that subsequent
processing can produce edible whey solids.
Whey processing required several modifications to ED systems. To retard
spoilage, the liquid streams had to be cooled. The inevitable accumulation of
solids on the membranes and spacers required routine cleaning in place and
periodic disassembly of the stacks for mechanical cleaning. These operations are
now well established and can be applied to the treatment of other food prod-
ucts.
LOCAL POLISHING CENTER

POINT OF USE
SPECIAL
FEE0
AESU cANlST.ERs_
-----==z====dl NUCLEAR GRADE
CANISTER
HP PUMPS A I LPP%S
NUCLEAR GRAD_ F ..1~6”
MiXF”
I
RESIN CANISTERS I
IO-20 PPM TOS P P
an-f-i
RAFWATEA / 1 1
__ I I
INLIZ,
200400
PPM TOS
ELECTROOIALYSIS UNITS STORAGE TANKS 1100 GALS EACH

CHARCOAL BEOS
Figure 8.6: integrated circuit process water system. (Bell Laboratories, Murray
Hill, NJ)
Cows’ milk is more salty than milk from human mothers, and this limits its
use in the preparation of infant formula. Desalting of cows’ milk by ED allows
larger quantities of cows’ milk solids to be used for these purposes. Research
has shown that desalting by ED to remove calcium improved the protein sta-
bility of frozen skim milk and its concentrates.’ The lumpy texture of thawed
frozen milk has been attributed to clumping of micellar casein, and calcium
removal led to the dissociation of micellar to serum casein.
The opposite effect was observed in blood ultrafiltrate. With the apparatus
shown in Figure 8.7, Jain* used ED to desalt the blood plasma. Chilling aided
in the precipitation of proteins which were removed by ultrafiltration. The salt
content of the plasma was restored by passing it through the concentrate com-
partments of the ED stack before returning it to the host. Such a process could
be used to treat certain diseases where excessive amounts of undesirable pro-
teins are present in a patient’s blood or to obtain needed proteins from a donor.
Other separations of proteins and amino acids have been achieved by pH
adjustment of the solution being treated. As in electrophoresis, a molecule that
is at its isoelectric point will not migrate in an electric field. However, a molecule
migrates as an anion at a pH above its isoelectric point, and as a cation below its
isoelectric point. Therefore, a solution of such a material at its isoelectric point
can be desalted by ED with minimal loss of the material.
Electrodialysis 493
Metal finishing processes offer numerous applications for ED in pollution

control and material recovery. The rinse streams from such processes pose par-
ticularly troublesome pollution problems. They are usually too dilute for direct
metal recovery and too concentrated for disposal. ED processing of a rinse
stream from a nickel electroplating system9 is illustrated in Figure 8.8. The used
rinse water flows through the depleting compartments of the ED stack where
the metal ions are transferred into the concentrate stream. The treated rinse
water can then be reused in the process. The concentrate stream can be recir-
culated to build up its metal content to a level that is useful for further recovery
or direct return to the plating bath.
ANTICOAGULANT
HOST UF ; - COOL -
I
1 I fro
A 1
I:
iii
0 I
’ HEAT - -IUF -
I
0 I
Figure 8.7: Membrane system for recovering proteins from blood.
RECOVERED
PLATED PARTS
a---
4 i---r
I t
RNSE TANK
L
RINSE
WASTE
Figure 8.8: ED for recovery of nickel from electroplating waste water.

ED has also been used to improve or maintain the quality of a plating bath
and thus eliminate the need for periodic replenishment. An instructive example
is shown in Figure 8.9 for the electroless plating of copper onto nonmetallic
substrates such as printed circuit boards. lo The process illustrated utilizes an un-
charged EDTA-CU complex where the reduction of copper is driven by the oxi-
dation of formaldehyde to formic acid. The formate ions and the sulfate ions
introduced with the make-up copper are undesirable reaction by-products that
are removed by ED. NaOH is added to adjust pH sufficiently to ionize the for-
mic acid. The uncharged EDTA-CU complex remains in the diluate stream, which
is returned to the plating tank.
A similar approach was used to recover silver from spent photographic
bleach-fixer solutions.” As shown in Figure 8.10, the iron, which was used to
oxidize the silver, remains in the bleach-fixer solution, because it forms a com-
plex with EDTA. The silver permeates the cation-exchange membranes to the
iron-free concentrate stream from whence it is reclaimed by electrodeposition
in a separate cell.
BIPOLAR MEMBRANES
Bipolar membranes consist of an anion-exchange membrane and a cation-

exchange membrane laminated together. Application of an electric potential of
about one volt to the bipolar membrane exposed to an electrolyte solution wil!
cause current flow, but the amount of current flow will be determined by the
orientation of the membrane. If the cation-exchange membrane faces the anode,
the current will be relatively high. If the anion-exchange membrane faces the
anode, the current will quickly drop to a low value as the anions and cations are
depleted from the membranes. Subsequent elevation of the potential in the lat-
ter case will increase the current which is carried by the only available ions, H+
and OH- ions generated at the junction by water splitting. This results in the
production of acidic and basic solutions at the surfaces of the bipolar mem-
branes, as illustrated in Figure 8.11. Multiple bipolar membranes can be placed
between a single pair of electrodes in an ED stack along with other ion-exchange
membranes for the production of acid and base from a neutral salt.
The current efficiency of acid/base generation and the purity of the acid
and base made with bipolar membranes drops off as concentrations increase, be-
cause Donnan exclusion diminishes with increasing solution concentrations. Fur-
ther, the production rate is limited by the rate of diffusion of water into the bi-
polar membrane. Nevertheless, there are substantial advantages to the process.
Since there are no gasesevolved at the bipolar membranes, the energy associated
with gas evolution is saved, and the power consumption is about half that of
electrolytic cells. Compared to the electrodes used in conventional electrolytic
cells, the bipolar membranes are inexpensive. Where dilute (e.g., 1 N) acids or
bases are needed, bipolar membranes offer the prospect of low cost and min-
imum unwanted by-products.
A commercially available ED stack with bipolar, anion and cation mem-
branes was used by the author to generate 1 N HCI and NaOH from NaCI. A cur-
rent density of 100 mA/cm’ was maintained with an applied potential of 2V/cell.
Electrodialysis 495
SPENT
la!
1 1 1 SOLUTION
Figure 8.9: Regeneration of chemical copper plating bath by electrodialysis.
T
I
tI t
tI c
I t
i ’ A
A
I
CU-EDTA
NapSO4
HCHO
NoC HO2
N.+ -
0 -0
a- - so;
a-
F1 ’ 1 - CHO;
T
~
eI ,
f
“2’
1 I
1
WASTE
1 ‘--=+jK~zzfO~
HiHd
Figure 8.10: Use of ED for regenerating spent photographic bleach-fixer solution.

o
t-
Figure 8.11: Bipolar membrane construction and operation,
The current efficiency was about BO%, and the purity of the acid and base ex-
ceeded 98%.
Several applications for bipolar membranes have been reported. Figure 8.12
shows a process for absorbing SOs from flue gas with a solution of NasSOs and
regenerating the solution with a water splitter containing bipolar membranes.12
The regeneration process is shown in F.igure 8.13. H+ ions generated in the bi-
polar membrane lower the pH of the solution and allow SO2 to be stripped out.
Na’ ions pass through the cation-exchange membrane into the basic solution.
The stripped acidic solution is combined with the basic solution and returned to
the absorber.
An ED stack with alternating bipolar and cation-exchange membranes (the
arrangement shown in Figure 8.13) was used to convert Na2C03 and/or NaHCOs
to more valuable NaOH with gaseous CO2 as the by-product.13 A stack with bi-
polar, anion-exchange, and cation-exchange membranes was used to convert
ethylenediamine dihydrochloride into ethylenediamine and HCI in separate
streams.14 As more bipolar membranes become available in commercial quan-
tities, more applications of bipolar membranes are sure to follow.
ELECTRODES AND STACK POWER
Electrode Reactions and Materials

An electric potential is the driving force for transport of ions in ED. This
potential is applied from an external power supply through electrodes situated
on either end of the stack of membranes and spacers. Current flow in the circuit
external to the stack is electronic, i.e., electrons flowing through metallic con-
ductors. However, current flow within the stack is electrolytic, i.e., ions flowing
through solutions. The transfer of electrical charge from electronic to electro-
lytic conduction is accomplished via electrode reactions.
Electrodialysis 497
. So2 DEPLETED
FLUE GAS
*REGENERATED
SO2
SO2 RICH ABSORBER ABSORBER
FLUE- LIQUOR
SPENT A;TK)RE;
WATER
SPLITTER so;JH2o STRIPPER PURGE SULFATE
Block diagram of SO2 recovery process.

(Allied Chemical Corp.)
Figure 8.12: Block diagram of SOs recovery process. (Allied)
SOLUTION
SO2 + H20 SOLUTION
Figure 8.13: Two compartment water splitting process for converting bisulfite
solutions to SOz and base.
Cathode Reactions
M+X + xe- + MO metal deposition
2H’+ 2e- + Hz (acidic solution) evolution of
2Hz0 + 2e- + H2 + 20H- (basic solution) 1 gaseous hydrogen
The first cathode reaction is common in electroplating but is undesirable

in ED. Evolution of Hz with the consumption of H’ or production of OH- ions
is common to all ED cathodes. These reactions are relatively mild compared to
the anode reactions, so electrodes that tolerate Hz and OH-are employed. Stain-
less steel is the most common cathode material.
Anode Reactions
MO --f M+X + xe- metal dissolution

2Hz0 -+ 0s + 4H+ + 4e- (acidic solution)’ evolution of
40H--+ O2 + 2Hz0 + 4e- (basic solution) > gaseous oxygen
2CI- + Cls + 2e- evolution of gaseous chlorine
MO +xOH- -+ M(OH)x +xe-
oxidation of electrode
2M” + 2xOH- + MsO, + xHzO + 2xe-
Anode reactions that produce dissolution or oxidation of the electrode

metal are avoided by selection of resistant metals such as platinum. The use of a
thin layer of platinum on a substrate of less expensive metal (e.g., titanium)
moderates the expenditures for anodes. However, titanium is not suitable for
reversible electrodes, because it is attacked by the Hz generated at the cathode.”
When the polarity of the electrodes must be reversed periodically, as in the EDR
process, platinized niobium has been used.
Some metals, e.g., lead and iron, will form stable conductive oxide coatings
under properly controlled anodic conditions.‘6”8 Metal electrodes with oxide
coatings are economical compared to those with platinum coatings, but they are
not reversible. Indeed, reversibility of ED electrodes is exceptional. Care must
be exercised to ensure that the polarity of ED electrodes is not inadvertently
reversed, because severe electrode deterioration can result.
Electrode Isolation
The reactions that occur at the electrodes are normally incidental and oc-
casionally detrimental to the desired separations that take place in the repeating
cell pairs of an ED stack. The H+ and OH- ions generated at the electrodes can
cause undesired reaction with the components of the feedwater and lead to
membrane fouling. Moreover, migration of some feed components into the elec-
trode rinse streams can cause precipitation on the electrodes or the spacers and
membranes close by. Therefore, it is prudent to carefully consider such possibili-
ties and isolate the electrodes to minimize undesired reactions.
Although the reactions at the cathode are not severe in terms of electrode
deterioration, fouling problems can be severe at the cathode. The presence of
pH-sensitive salts, e.g., Ca(HC0-J2 in the feedwater is a matter of concern, be-
cause the OH- ions generated at the cathode can lead to precipitation.
Ca(HCOs)s + 20H- + CaC03 + COT + 2HsO
The problem of pH control in the cathode rinse stream (catholyte) is usu-

ally handled by acidification. Faraday’s constant, 96,500 A-sec/eq, is used to cal-
culate the amount of acid required to neutralize the OH- ions generated at the
cathode. While acid addition contributes significantly to operating costs of com-
mercial ED plants, the fact that there are hundreds of cell pairs between each
pair of electrodes keeps the catholyte acid costs at tolerable levels. In contrast,
acidification of the feed stream to prevent precipitation can increase operating
costs substar,tially.
Electrodialysis 499
Selection of the type of membrane to be used at the cathode end of the

stack is important. If a cation-exchange membrane is used, the catholyte is an
enriching stream. This means that cations from the feed solution can enter the
catholyte. The use of a univalent-cation-selective membrane next to the cathode
can alleviate some of the problems of multivalent cations in the catholyte. An
enriching catholyte stream is almost invariably a waste stream, which presents
disposal problems. Moreover, makeup water of good quality is required. The
overriding advantage of the cation-exchange membrane is that it blocks the entry
of cathode-generated OH- ions into adjacent compartments.
The catholyte is a depleting stream if an anion-exchange membrane is placed
next to the cathode. The major advantage to this arrangement is that cations
from the feed solution can be almost completely excluded from the catholyte.
The price to be paid is that acid must be added to the catholyte in sufficient
quantity to supply all of the anions needed to carry current through the mem-
brane.
The major considerations about isolation of the anode pertain to the acidic,
oxidizing conditions there. An anion exchange membrane is often used to block
the passage of H’ ions into the rest of the stack. However, anion exchange mem-
branes are only moderately effective at blocking H+ ions, so neutralization of the
anolyte may be needed if acid-sensitive components, e.g., proteins, are present
in the feed. The undesirable formation of hypochlorite occurs at the anode
whenever chloride ions are present in the anolyte. The use of a cation-exchange
membrane at the anode prevents chloride entry from the feed solution to the
anolyte, but this arrangement makes the anolyte a depleting stream and requires
a supply of electolyte (e.g., NaOH, Na2S04) for the anolyte.
In some circumstances, a common rinse solution can be circulated through
the anode and cathode compartments. The simplicity and economy of a single
electrode stream must be weighed against the problems of current leakage through
the solution lines, buildup of undesirable materials such as hypochlorite, and the
mixing of H, and O2 generated at the electrodes. These problems may be con-
veniently ignored in laboratory experiments, but they are critical issues in large
ED stacks.
Kedem et al demonstrated that gas evolution could be avoided by the use
of powdered activated carbon in a common electrode rinse stream.lg The large
surface of a carbon particle became charged when it came in contact with the
anode or with another positively charged carbon particle. This charge was neu-
tralized by sorption of anions from the rinse solution. When the suspension
reached the cathode, the charge on the particle was reversed and the sorbed
anions were exchanged for cations. With a 2.5% suspension of carbon in 0.02
to 0.2 normal solution, a current of lo-20 mA/cm’ could be sustained for weeks
with no gas evolution. ,,
The benefits of ,a common electrode rinse stream were demonstrated in a
photographic devel per regeneration process illustrated in Figure 8.14. The use
of cation-exchan $ membranes at both electrodes made the anolyte a depleting
stream and the catholyte an enriching stream. This kept the 8r- out of the elec-
trode rinse solution and conserved the Na&Os that maintained the conductivity
in the electrode compartments.20 A similar arrangement with NaOH solution for
the electrode rinse allowed the use of an inexpensive ferrous anode in an ED
stack.”
DEVELOPER
Figure 8.14: The use of cation-exchange membranes at both electrodes conserved

the Naz CO3 in the common electrode rinse stream of a photographic developer
regeneration process.
Power Supplies
Ionic transport through the solutions and membrane of an ED stack is
driven by a direct electric potential. The components required to safely deliver
current (DC) at the proper voltage are illustrated in Figure 8.15. An isolation
transformer contributes measurably to the safety of operating an ED stack, es-
pecially an experimental one from which samples are being taken. The output
from the isolation transformer has a floating ground if the load is not inten-
tionally grounded. If the circuit is accidentally grounded by an operator’s
hand, this contact point becomes the zero potential point, but no shock is ex-
perienced.
The power level of the ED stack is regulated by adjustments of the stack
voltage. In laboratory ED units an autotransformer or a solid state voltage regu-
lator is preferred, because the user will likely want to experiment with the ef-
fects of applied voltage on stack performance. In large installations a trans-
former may be used, or the number of cell pairs in the stack may be varied to
match the desired cell-pair voltage drop to the available DC voltage. Since
transformers operate on AC, voltage adjustment must occur before conversion
to DC.
A full-wave rectifier is commonly used to convert AC to DC. The current
from such a rectifier has a 120 Hz ripple which is of no consequence. The rates
of build-up and collapse of concentration boundary layers are so slow compared
to the ripple frequency that stack performance is unaffected by the ripple.
Electrodialysis 501
LJJ
A-C SOURCE
ISOLATION TRANSFORMER
AUTOTRANSFORMER
FULL-WAVE RECTIFIER
METERS
DRY CONNECTIONS TO
ELECTRODES
Figure 8.15: Power Supply for ED.
Connections between the power supply and the electrodes are usually made
outside the stack so that they can be kept clean and dry. It is advisable that the
conductor between this connection and the electrode be of the same material as
the electrode and that the connection with the electrode be welded. Otherwise,
severe corrosion problems can result.
The observation that the cell-pair resistance is reasonably constant (see
Figure 8.16) over the range of safe operation of ED allows one to develop a
simplified equation concerning power consumption in an ED stack. With con-
stant R,, the cell voltage is proportional to cell current, i.e., E,, = IR,,.
The typical voltage drop across a cell pair is in the range of 1 to 2 volts. E,,
does not change greatly as feed concentrations change, because I is proportional
to concentration, and R,, varies inversely with concentration.
Assuming constant current efficiency, the current is proportional to the
amount of salt removed by the passage of current, i.e., I a C. Since power con-
sumption is the product of current and voltage, one reaches the not-too-surpris-
150,
FF IV STACK ASSEMBLED
WlTH TOKUYAHA “EHBRANES
140 - AND 23-MIL VEXAR SPACERS
VELOCITY - 19.6 CM/SEC
-= 130 - FEED - 0.0167 E ~aC1
” LIMITING VALUE OF i/y TAKEN
COULOMB EFFICIENCY -
FROU COWAN-BRDUN PLOT
0.94-0.97
g 120 -
.
z 110 -
100 -
B-O-
I
1
I
I
I
I
7.0 - 0 I..
z * -1
I
6.0 -
C. , I 1
0 200 400 600 900 1000 1200 1400 16 00
i/E, * OR AMP CMIEQ
Figure 8.18: Effects of i/N on cell-pair resistance (R,P) and pH of the depleted
effluent.
ing conclusion that the power consumption in the ED stack is proportional to

the amount of salt removed, i.e., P a AC. Reductions in power consumption
can be achieved for any ED stack by reducing the applied voltage, but a conse-
quent reduction in electrolyte removal would occur.
CONCENTRATION POLARIZATION
The phenomenon of concentration polarization is important in electro-

dialysis, because it ultimately limits the rate at which ion transport can occur.
Concentration polarization will be discussed first in a descriptive manner. Then,
a simplified mathematical model will be used to show the quantitative aspects
of concentrations polarization and explain the important concept of “limiting
current density”.
In Figure 8.1, it was shown that the interposition of pairs of anion- and
cation-exchange membranes between electrodes resulted in the depletion or
enrichment of electrolyte in the alternating solution compartments when an
electric potential was applied. Close examination of the system would reveal non-
uniformity of electrolyte concentration within a solution compartment. In the
enriching compartments, the concentration is higher at the membrane surfaces
Electrodialysis 503
than in the bulk solution, and in the depleting compartments, the concentra-
tion is lower at the membrane surfaces. The resistance to the flow of electric
current increases as this interfacial concentration decreases. The reason for the
nonuniformity of concentrations in the solution compartments is that cations
carry virtually all of the electric current in the cation exchange membranes but
only about half of the current in the solution.
The fraction of the total current carried by a particular ionic species in an
electrolyte solution is called the transport number t. For a solution containing
a univalent electrolyte like NaCl or KNOa the transport number of the cations is:
+
t = A +/( x++x-)
where h is the equivalent ionic conductance. Values of h extrapolated to infinite
dilution are shown in Table 8.2. Solutions with more than one salt or with multi-
valent ions require a more generalized equation to describe transport numbers:
= xizi2ci/l: 1 izi2ci
ti
where Ci is the molar concentration and Zi is the ionic charge of each species
present in the solution.
Transport numbers within the membranes are not as easily determined as
those in solutions because of interactions between the mobile ions and the fixed
ionic groups. Although the generalized equation above is applicable to the mem-
brane as well as to the solution phase, the values of h and C are different in the
membrane. Because of Donnan exclusion, the concentration of counter ions
(those of opposite charge to the fixed ions) is always much higher than the con-
centration of co-ions (several orders of magnitude higher when external solutions
are dilute). Since most of the ions available to carry current within the mem-
brane are counter ions, the transport numbers of counter ions are high, typically
above 0.9 when external solutions are dilute (<l molar).
What happens with increasing ED current can be presented mathematically
with the aid of a simplified model called the Nernst idealization. This model,
which is illustrated in Figure 8.17, is based on the simplifying assumptions of
stagnant boundary layers of constant thickness and a well-mixed region in the
center of the solution compartment. These idealized boundary-layer conditions
are obviously unrealistic, especially in the presence of the spacer screens used in
an ED stack, but they allow a simplified approach to an otherwise complex
problem.
Now, consider the region of the system designated by the square in Figure
8.17 for a single uni-univalent electrolyte such as NaCI. The flow of electrical
charge in this region is designated by the symbol i with the units mA/cm’. With
the additional assumption that diffusion is negligible within the membrane, flux
of cations through the cation exchange membrane is proportional to the electric
current. - -
Je+ = i t+/F
Here the overbar designates the membrane phase, and i’ is the transport
number for cations in cation exchange membrane. The F is Faraday’s constant,
Table 8.2: Equivalent Conductance in Aqueous Solution
II+ 316.55 349.9
L1+ 33.29 39.69
N8+ 43.39 50.11
r+ 40.4 64.44 73.52
ilb+ 67.5
C.+ 44.4 67.65 79.1
M+ 32.9 53.77 61.92

Tl+ 65.1 74.7
Nit: 73.4
iCrE.1d’ 29.1
VIP- 53.06
m++ 59.50
4sr++ 59.46
4Ba* 63.64
Ku++ 46.3
42n++ 46
OH- 176.6 197.6
?- 46.65
Cl' 41.3 65.41 76.34

Rr- 43.1 67.4 79.3
I- 43.4 67.4 16.9

c10; 54.99
_
Bra, 47.9
10; 34.02
NO; 40.4 61.74 71.44
c10: 59.4 69.0
MO; 31.0 61.7
lICoo- 47.0
CB,CCO- 20.3 33.5 40.9
Piera- 25.4
WOr- 69 79
41
39 63 73
‘C. W. Davhs, %oductivity of solutiomr Chapan

and Ball, 1930.
by. S. uuned and 8. I).Cum, lEloctrochmiary of
solution8,~ Roinkwld, 1950.
Electrodialysis 505
,.- Static N._
-i-i- -i-T
6 = Static boundary-layers
C = Cation-exchangemembrane
A = Anion-exchangemembrane
Figure 8.17: Concentration gradients in electrodialysis.
96,500 A-sec/eq. Since i’ = 1 in the ideal case, one equivalent of cations is re-
moved through the membrane for every faraday of current flowing through the
region of interest. Now in the aqueous phase near the membrane, the flux of
cations transported electrically is:
Je+ = i t+/F
If the electrolyte were NaCl where-t’ = 0.4, the electric current would deliver
only 0.4 of an equivalent of cations for each equivalent of cations transported
through the cation exchange membrane. The remaining 0.6 faraday of current
would be transported by anions that are moving in the opposite direction. The
net result would be a depletion of anions and cations from the solution at the
membrane surface. This depletion would result in a concentration gradient that
would allow diffusive transport of the balance of ions needed for electroneutral-
ity. In this case, the balance would be 0.6 of an equivalent per faraday of current
flow. The diffusive flux of ion pairs down this linear concentration gradient is:
= olc,-ci 1/ 6
Jd
The subscripts b and i refer to the bulk and interfacial concentrations. Now,
one can write the steady-state flux equation for cations in the vicinity of the
membrane solution interface.
$+ = Jet + Jd
i?/F = it+/F + D(Cb-Ci)/ 6

Rearrangement yields:
i/(Cb-Ci 1 = DF/6 (?+-t+)
F is a constant and D, 6, t’ and t* are essentially nonvariant for a given set

of operating conditions (temperature, flow rate, membrane and salt type). Now,
i is an independent variable that can be adjusted by changing the voltage applied
to the electrode. When i changes, the value of Cb-Ci changes proportionately.
However, there is a limitation on the extent to which the value of i in this equa-
tion can be increased. That limit is approached as Ci approaches zero. The limit-
ing current density can thus be expressed:
= DF/ 6 (t+-t+)
i I im’Cb
For a solution like KCI where the ionic conductances of both ions are the
same, t+ = t- and irim is the same value for both interfaces. However, this is not
the case for NaCl where t+ = 0.4 and t-= 0.6. The equations above indicate that
the value of ilk,, at the anion exchange membrane would be 1.5 times higher
than that at the cation exchange membrane.
The data in Figure 8.16 show the results of an experiment to determine the
values of irim. Those data were obtained from electrodialysis of a solution of
NaCl in a stack containing ten cell pairs.21 The applied voltage, current, and pH
of the depleted product water were monitored. A control experiment with one
cell pair in the stack was used to determine the voltage drop attributable to the
electrodes and rinse streams. The cell-pair resistance, R,, = (V,,,,r - V,r,) x A,,,/l,
was observed to increase slightly for moderate increases in the polarization
parameter, i/N. [The polarization parameter is defined as the current den-
sity, i = I/A,n, divided by the log-mean concentration of the depleting stream,
N = (Ci, - Cd,t)/ln(Ci,/Co,t).l
However, the value of R,, increased more rapidly when the value of i/N
exceeded 950 A-cm/eq, indicating ir,,,, at the cation exchange membrane. More-
over, a slight increase in the pH of the depleted effluent was detected. When the
value of i/N reached about 1,400 A-cm/eq, the pH of the depleted effluent
dropped precipitously, indicating il.,,,, at the anion exchange membrane. The
ratio of the irim values was 1.5 as expected.
Upon close examination of the data in Figure 8.16, one might question
whether the inflection points in the curves truly represent the limiting values of
the polarization parameter, i/N, i.e., whether the concentration approaches zero
in the solution at the interface of the cation exchange membrane. If the con-
centration were to reach zero over the entire surface at the same value of i/N,
the inflection would likely be more abrupt and the curve would become steeper,
because splitting water to form H+ ions would be the only source of ions to carry
the additional current. In fact, the slight increase in pH of the depleting solution
suggests that water splitting was not significant. A more reasonable explanation
is that localized depletion occurred at the inflection point and expanded as cur-
rent increased.
The maximum point in the pH curve was attributed to attainment of the
limiting value of i/N at the anion exchange membrane. Again this was probably
localized depletion that expanded as current increased, but the pH decrease was
much more dramatic than the pH increase observed when polarization occurred
Electrodialysis 507
at the cation exchange membrane. This observation indicated that water splitting
occurred at the anion exchange membrane but not at the cation exchange mem-
brane. The prevailing theory explains that the kinetics of water splitting in solu-
tion are too slow to cause appreciable pH changes to occur.22 Instead substantial
water splitting occurs in the anion exchange membrane, and it is catalyzed by
tertiary amines that withdraw protons from water molecules.
The current density at which R,, begins to rise rapidly is generally referred
to as the limiting current density. This terminology obviously does not refer to
a physical limitation to the current flow but rather to a practical limitation for
trouble-free operation of electrodialysis. Moreover, the precise value of irim is
not as easily established as Figure 8.16 would indicate. Uneven flow distribution
within or between solution compartments could lead to localized thick bound-
ary layers that result in lower values of ijim. Also, the presence of non-conduc-
tive spacers in the solution compartments causes local current densities to be
higher than the calculated average value. Therefore, for a margin of safety, one
would operate an electrodialysis stack at a current density substantially below
the measured limiting value.
MEMBRANE FOULING
When the feed solutions to ED stacks are clean and are relatively free of
sparingly soluble materials, ED stacks can be operated for years with little con-
cern about membrane fouling. However, most feedwaters have constituents that
can cause problems for ED stacks. Such problems are usually handled by reduc-
ing current densities, periodic current reversal (the EDR process), or pretreat-
ment of the feedwater.
Figure 8.18 illustrates some of the fouling problems. Large organic mole-
cules with ionizable groups (e.g., humic acids from decomposing vegetation) are
troublesome foulants for ED membranes. Their negative charges allow them to
migrate in the electric field but their large size prevents their passage through
the membrane. Consequently, these charged molecules tend to accumulate on
the dilute side of anion exchange membranes and result in a buildup of mem-
brane resistance. Colloidal particles, which are also usually negatively charged,
cause similar problems.
Sparingly soluble salts such as CaS04 and CaC03 can precipitate on the con-
centrate side of the membrane, because their concentrations are highest there
due to concentration polarization. CaCOs precipitation occurs when the limiting
current density is exceeded. The reaction of HCOs- ions with OH- ions generated
by water splitting forms CO; ions that pass through the anion exchange mem-
brane and precipitate in the boundary where the concentration of Caf+ ions is
the highest.
Problems of membrane fouling are generally less severe when current den-
sities are kept low to minimize concentration polarization. CaS04 precipitation
is countered by limiting the concentration in the enriching stream below the
saturation level, and by addition of sodium hexametaphosphate to delay crystal
formation. CaCOs precipitation can be prevented by acidification of the enrich-
ing-steam feed. Fouling of membranes by organic anions and colloidal material
- +I
+
HCO; + Ca++
+
+
+
+
+
+I
d
Figure 8.18: Large organic ions and pH-sensitive salts can foul anion-exchange
membranes.
is best handled by pretreatment to eliminate the offending material. Alternat-

tively, these membrane-fouling problems can be alleviated to some extent by use
of the EDR process.
EDR is the electrodialysis reversal process developed by lonics, lnc.23 The
EDR process employs reversible electrodes and automatic valves to swap the
flows of enriching and depleting product lines. The reversal is programmed to
take place about three times per hour. This frequent reversal tends to dislodge
the organic and particulate foulants from the membrane surfaces. Moreover,
since current reversal causes concentrating boundary layers to become depleting
boundary layers, freshly formed precipitates tend to dissolve. Immediately after
the current reversal the product water is of poor quality due to the discharge
of the aforementioned foulants. Therefore, there is a period of a few minutes
when the product water must be diverted to waste. This loss of feedwater along
with the high costs associated with the automatic switching valves are the major
drawbacks to EDR. The proponents of EDR23 claim that these disadvantages
are more than offset by increased membrane life, reduced chemical costs, higher
operating currents, and less maintenance than with conventional ED. Super-
saturation of CaS04 up to 220% in the brine stream was tolerated by EDR. The
supersaturation was pushed to 440% with the addition of acid and sodium hexa-
metaphosphate.
Although polarity reversal has been lauded as a major breakthrough by
lonics, Inc., it has not been adopted by other suppliers of ED equipment. This
avoidance of polarity reversal does not appear to be due to strong EDR pat-
ents, but rather to a general impression that removing foulants by pretreatment
is more cost effective than attempts to moderate their damage to the mem-
branes.” In fact, the extent of pretreatment in EDR plants would likely make
the water acceptable for conventional ED.24
Electrodialysis 509
COST OF ED
The costs of constructing and operating an ED system are dictated by the

quality of the water to be treated and the degree of desalting that must be
achieved.25 There are economies of scale-up in all of the costs except energy and
chemicals. The costs are generally divided into capital, fixed operating, and var-
iable operating categories. Capital cost include the site, building, stacks, tanks,,
piping, pumps, power supplies, and pretreatment equipment. Capital costs are
usually above $l/gpd capacity except for very large plants. These costs can be
amortized over the expected life of the system and expressed in terms of prod-
uct output, e.g., $/kgal. These cost begins at about $l/kgal for large desalting
plants for desalting brackish water to drinking water standards, but they can
reach several dollars per kgal for small installations. Electrical power consump
tion varies with the salinity of the feedwater and the applied voltage. Except
for seawater where consumption of 28 wh/gal was reported,5 power costs are
usually in the range of $0.50 to $l.OO/kgal.
The costs of desalination by ED and RO are very close. ED is generally the
less expensive proces? where the salinity is low, because the current is propor-
tional to the amount of salinity reduction. RO is considered to be less expensive
for seawater, but low-resistance membranes and thinner solution compartments
could make ED competitive for seawater desalting.
REFERENCES
1. Helfferich, F., /on Exchange, New York: McGraw-Hill (1962).

2. Kneifel, K. and Hattenbach, K., Desalination 34: 77-95 (1980).
3. Trivijitkasem, P. and Ostvold, T., Water transport in ion exchange mem-
branes. Electrochim. Acta 25: 271 (1980).
4. Zmolek, C.R., Ind. Water Eng. Dee: 6-11 (1977).
5. Tani, Y., Energy-saving electrodialyzer for seawater desalination. Tech.
Proc. - Annu. Conf. Int. Trade Fair Nat’l. Water Supply Improv. Assoc.
9th. Paper Number 4 (1982).
6. Kawate, H., Seto, T., Komori, R., and Nagasato, Y., Symp. Salt. Proc. 5th:
317-24 (1980).
7. Lonergan, D.A., Fennema, 0. and Amundson, C.H., J. Food&i. 47: 1429-34
(1982).
8. Jain, S.M., U.S. Patent 4,275,140 (1981).
9. Itoi, S., Nakamura, I. and Kawahara, T., Desalination 32: 383-9 (1980).
IO. Grenda, D.W., European Patent 0015737 (1980).
11. Ono, T. and Watanabe, M., U.S. Patent 4,204,930 (1980).
12. Liu, K.J., Chlanda, F.P. and Nagasubramanian, K., J. Membr. Sci. 3: 57-70
(1978).
13. Gancy, A.B. and Jenczewski, T.J., U.S. Patent 4,238,305 (1980).
14. Chang, V., J. Appl. Electrochem. 9: 731-6 (1979).
15. Foster, P.A., Rowe, M.J. and Farrar, J.B. ED: Polarity reversal or not? Paper
presented at 46th International Water Conference, Pittsburgh, Pennsyl-
vania (November 1985).
16. Pavlov, D. and Dinev, Z., J. Nectrochem. Sot. 127 (4): 855-63 (1980).
17. Carr, J.P. and Hampson, N.A., Chem. Rev. 72 (6): 679-703 (1972).
18. Tejeda, A.R., U.S. Patent 3,869,364 (1975).
19. Kedem, O., Cohen, J., Warshawsky, A. and Kahana, N., Desalination 46:
291-299 (1983).
27. Yamada, M., Matshshita, S., Hirai, H., Tsuyuki, I. and Ogawa, Y., J. Appl.
Photog. Eng. 7 (2): 53-8 (1981).
21. Davis, T.A. and Lacey, R.E., Forced-flow electrodesalination. OSW R&D
Progress Report No. 557 (1970).
22. Rubinstein, I., Warshawski, A., Schechman, W. and Kedem, O., Desalina-
tion 51: 55-60 ( 1984).
23. Katz, W.E., Proc. lnt. Water Conf. Eng. So., West Pa. 44: 51-62 (1983).
24. Mansouri, M., Electrodialysis reversal units used as predemineralizer in boiler
feedwater treatment. Paper presented at 45th International Water Con-
ference, Pittsburgh, Pennsylvania (October 1984).
25. Reed, S.A., Desalting seawater and brackish waters: 1981 cost update.
ORNL/TM-8191 (1982).
26. Schmoldt, H., Strathmann, H. and Kaschemekat, J., Desalination 38:
567-82 (I 981).
9
Coupled Transport Membranes
Richard Baker and lngo Blume
INTRODUCTION
Coupled transport is a membrane process for concentrating ions and separat-

ing ions from aqueous solutions. The membrane used in this process consists of a
water-insoluble liquid containing an ion-complexing agent that is specific for the
ion of interest. The desired ion is complexed at one interface of the membrane,
forming a neutral-ion complex. The neutral ion-complex then diffuses across the
membrane to the opposite interface, where the reaction is reversed by making
appropriate changes in the external solution conditions. The reformed complex-
ing agent then diffuses back across the membrane, where it picks up more of
the desired ion. Thus, the complexing agent acts as a shuttle to carry ions across
the membrane.
The coupled transport process is illustrated in Figure 9.1 for Cu2+ and H+
ions and a complexing agent denoted by RH. The water-immiscible agent fills
the pores of a microporous membrane, thus forming a liquid organic membrane.
The equilibrium that exists at the two membrane-solution interfaces is:
2RH + Cu2+ _ CuR2 + 2H+
At the feed solution interface (i.e., the left interface in Figure 9.1) , the copper
ion reacts with agent, liberating two hydrogen ions. The neutral copper complex
CuRz is soluble only in the organic phase, ano it diffuses across the membrane
to the product solution interface, which is maintained at a lower pH. There, the
reaction is reversed, liberating the copper ion and consuming two hydrogen ions.
This step reforms the neutral complexing agent, which then diffuses back across
the membrane. Thus, copper ions move from left to right, and the movement of
hydrogen ions in the opposite direction maintains electrical neutrality. It is this
coupling between the flow of one species (CL?‘) and the other (H+) that gives
coupled transport its name.
511
Carrier
Membrane
Cu*++ SHR-CUR +2H+f z CuR 2

+2H+-Cu*++2HR
2
Figure 9.1: Coupled transport scheme for copper.
Proper selection of complexing agent and conditions makes clean separa-

tions of metal ions possible. Further, ions of interest can be concentrated against
their concentration gradients. Coupled transport membranes can thus be consid-
ered as chemical pumps. The energy for the pumping action derives from the
flow of one species (hydrogen ion in the example shown here) down its concen-
tration gradient.
Coupled transport is not limited to cations. Anions such as Cr207*- can be
transported by a completely analogous processes using, for example, tertiary
amine complexing agents.
HISTORY AND BACKGROUND
The historical development of coupled transport is shown schematically in

Figure 9.2.’ This process originated in early experiments of biologists using nat-
ural carriers contained in cell walls. As early as 1890, Pfeffer postulated trans-
port properties in membranes using carriers. ’ Perhaps the first coupled trans-
port experiment was performed by Osterhout, who studied the transport of am-
monia across algae cell walls.3 By the 1950’s, the carrier concept was well de-
veloped, and workers began to develop synthetic biomembranes analogues of
the natural systems. For example, in the mid-1960’s, Sollner and Shean4” stud-
ied a number of coupled transport systems using inverted U-tubes, shown in
Figure 9.3. At the same time, Bloch and Vofsi published the first of several
papers in which coupled transport was applied to hydrometallurgical separations,
namely the separation of uranium using phosphate esters7-” Because phosphate
esters were also plasticizers for polyvinyl chloride (PVC), Bloch and Vofsi pre-
pared immobilized liquid films by dissolving the esters in a PVC matrix. Typi-
cally, the PVC/ester film was cast on a paper support. Researchers actively pur-
sued this work until the late 1960’s. At that time, interest in this approach
lagged, apparently because the fluxes obtained did not make the process com-
petitive with conventional separation processes. Some workers are continuing to
apply these membranes to metal separations,“,‘* but most current interest in
PVC matrix membranes is in their use in ion selective membrane electrodes.‘3
Coupled Transport Membranes 513
,983 -rirrt pilot plant Installed

in *yonsng uranium mill
Figure 9.2: Historical development of coupled transport membranes.
Organic Carrier Phase
Feed Solulion
' pt0duct Solutlon
Figure 9.3: Early U-tube liquid membrane system.

Following the work of Bloch and Vofsi, two other methods of producing
immobilized liquid films were introduced. Both are still under development. In
the first approach, the liquid carrier phase is held by capillarity within the pores
of a microporous substrate, as shown in Figure 9.4. This approach was first used
by Miyauchi14 and further developed by Baker et al15”7 and by Largman and
Sifniades.” The principal objective of this early work was the recovery of copper
and other metals from hydrometallurgical solutions. Despite considerable effort
on the laboratory scale, the first pilot plant was not installed until 1980.‘9”o
The principal problem is instability of the liquid carrier phase in the micro-
porous membrane support.
Feed Solution Product Solution
Organic
Carrier
Microporous
Polymeric Phase
Membrane
Figure 9.4: Supported liquid membrane.
The second type of immobilized liquid carrier is the emulsion or “bubble”

membrane. In this technique, a surfactant-stabilized emulsion is produced as
shown in Figure 9.5. The organic carrier phase forms the wall of the emulsion
droplet separating the aqueous feed from the aqueous product solutions. Metal
ions are concentrated in the interior of the droplets. When sufficient metal has
been extracted, emulsion droplets are separated from the feed and the emulsion
is broken, liberating a concentrated product solution and an organic carrier
phase. The carrier phase is decanted from the product solution and recycled to
make more emulsion droplets. The principal technical problem is the stability
of the liquid membrane. Ideally, the emulsion membranes must be completely
stable during the extraction step, to prevent the two aqueous phases mixing,
but must be completely broken and easily separated in the stripping step. Achiev-
ing this level of control over emulsion stability has proved difficult.
The technique of emulsion membranes was popularized and fully developed
by Li and his coworkers at Exxon.2’-2g Starting in the late 1950’s and continuing
for more than twenty years, the Exxon group’s work led to the installation of
the first pilot plant in 1979.29 Although the process is still not commercial, a
number of pilot plants have been installed, principally on hydrometallurgical
feed streams. Another important group working independently on the problem

at about the same time was Cussler, Evans and others at Carnegie Mellon.30-34
More recent workers in the field include Halwachs and Schuger3sJ8 in West
Germany, Marr and Kapp in Austria,39t40 Stelmaszek et al in Poland,41r42 Martin
and Davis43 in the United Kingdom, and Danesi et al,44t4s Noble et al,46f47 and
Lamb, Christensen and lzatt48”’ in the United States.
A number of reviews on various aspects of liquid membranes and coupled
transport processes have also appeared.s2”s
Organic
Product Solution
Carrier
Phase
_ _(..(,,,. _ lrJl
... ” \. _ .,, . I .<,s,aa..
\
Feed Solution
Figure 9.5: Surfactant stabilized emulsion membrane.
COUNTER TRANSPORT AND CO-TRANSPORT
Coupled transport processes can be divided into two categories, depending

on the type of reaction occurring between complexing agent and permeant. The
first type is called counter transport (shown in Figure 9.6). The key feature of
counter transport is that the fluxes of the two permeating ions move counter to
each other across the membrane. The reaction in this case is:
2RH + Cu2+ Z=? CuR2 + 2H+ (2)
More generally, the reaction can be written
nRA + M _ MRn + nA (3)
In the reaction for the transport of copper, M is Cu*+ and A is H’. Equation 3
does not have to be limited to cation transport. For example, consider the trans-
port of uranyl ions by the tertiary amine salt (R3NH)*S04,
(R3NH)2S04 + Hg2(Sg4)2 --(R3NH)2U02(Sg4)2 + Sg42- (4)
In this case, M is UO2(SO4)*- and A is Sod*-.

Membrane
nRA + M--MRn + nA MRn + nA-+nRA + M
M
---w
A
MR
/
I
Dilute Concentrated
A and M
/ A and M
/AR
\
M
--m-
A
I
t
M
4
A
Figure 9.6: Counter-coupled transport.
The second type of coupled transport is co-transport, illustrated in Figure

9.7. The key feature of co-transport is that the fluxes of the two permeating ions
move in the same direction across the membrane. The general form of the reac-
tion in this case is:
A+N+R -RAM (5)
A typical example of this type of process might be the transport of uranyl ions
by tertiary amine complexing agents via the reaction:
4R3N + 4H+ + 002(SO4)3 ‘1’=(R3NH)4 002(SO4)3 (6)

In this example, M is UOz(SO&4- and A is 4H’. Another example of co-trans-

port is the transport of alkali metals by crown ethers by reactions of the type:
Cl- + Na+ + Crown -Crown (NaCl) (7)
where Crown is a cyclic ether.”
Membrane
A + M + R-rRAM RAM-A + M + R
I I I l
t
M
t
A
Figure 9.7: Co-coupled transport.
THEORY
Several authors have attempted to provide a general theory of coupled trans-

port and the closely related process of facilitated transport, including Smith
et al,56 Schultz et al5749 Cussler5’ and Danesi.44r45 The usual approach is to as-
sume an equilibrium between the various reacting species in the membrane phase
and link these concentrations by the appropriate equilibrium constants. A dif-
fusion model using Fick’s law experiments is then set up for each species and
these expressions are integrated between the boundary values defined by the
membrane distribution coefficients and the permeant concentrations in the
surrounding aqueous solutions. The result is an expression for the permeant flux
that includes terms for permeants diffusion coefficients, equilibrium constants,
and distribution coefficients. These expressions are complex and of little prac-
tical value.
A simpler approach others have followed15’16!60 is to ignore the effect of
minor species in the membrane and aqueous phases. For example, consider the
counter transport reaction:
2RH + Cu2+4 *CuR2 + 2H+ .

In principal, all four species are in equilibrium in the aqueous solution and in
the organic membrane phase, but in practice CuRz and RH are confined to the
organic membrane phase and Cu*+ and 2H’ are confined to the aqueous phase.
Equilibrium between the reactants exists only at the membrane interface. We
therefore assume that concentrations of the permeants at each interface to be
defined by an appropriate equilibrium constant and then substitute these values
in a simple Fick’s law expression.
Consider the general case of a permeant M of valence n reacting with car-
rier RA to form the complex MR, inside the liquid membrane:
nRA + fl _ MRn+nA . (8)
This reaction is characterized by an equilibrium constant:
CMRnI
[Al”
K (9)
= [RA]” [M]
This equilibrium constant could be written for either the organic phase or the
aqueous phase. However, only [MR,] and [RA] are measurable in the organic
phase, where [A] and [Ml are negligibly small. Similarly, only [A] and [Ml are
measurable in the aqueous phase, where [MR,] and [RA] are negligibly small.
We can, therefore, write Equation 9 as:
[ MRn] ’ [A] ‘In

K’ = km . K,
(IO)
[RA]‘” [Ml” = k,
where the superscript ’ refers to the organic phase, superscript ” refers to the
aqueous phase, and k, and k, are the partition coefficients of M and A between
the aqueous and organic phases. We prefer this formulation of Equation 10 be-
cause all the quantities are easily accessible experimentally. For example,
[M&l '/D'Jl ” is easily recognizable as the distribution coefficient of metal
between the organic and aqueous phases.
The same equilibrium constant must apply at both membrane-solution
interfaces, and we can recast Equation IO into the following form:
CMRnIi[All” CMRnIi
CA]&” (11)
K’ =
[RA]$-’ [Fl]b: = [RA-jkn [M];
where o and II refer to the two sides of the membrane.

Consider now the situation where a counter ion concentration gradient is
established that exactly balances the metal ion concentration gradient, so there
is no flux of either ion across the membrane. Under this condition, [MR,],!, =
[MR,] b and [RAI b” = [ RA] h”, and we obtain the expression:
(12)
Thus, the maximum concentration factor of metal ion that can be established
across the membrane varies with the counter ion concentration ratio (in the
same direction) raised to the nth power.
This development, of course, demonstrates nothing about the metal ion
flux across the membrane under non-equilibrium situations. For this, we turn
to Fick’s law. At steady state, the flux JMR,,, in mol/cm2-see, of metal complex
MR, across the liquid membrane is given by:
6’lRn ([MRn]; _ CMRnIl) (13)

JMR, =
II
where DMRn is the mean diffusion coefficient of the complex in the membrane
of thickness Iz. In principal, JMR ,, can be measured experimentally in a permea-
tion experiment, and the values of [MR,] c_anbe determined in equilibrium dis-
tribution coefficient measurements. Thus, DMR n can be measured as a function
of concentration, pH, and other conditions. However, this approach demon-
strates nothing about coupling.
To illustrate coupling effects, we begin with Equation 13 and eliminate the
terms in [MR,l by introducing Equation 10. This results in a complex expres-
sion involving the desired quantities [Ml and [A], but also involving [RA] .
However, from mass balance considerations, we have the following relationship:
n[MRn] + [RAI = Y 9 (14)
where y is the molar density of pure RA. We have made the reasonable assump-
tion here that there is no volume change in the interconversion of n mols of RA
to n mols of MR,.
We can now substitute Equations IO and 11 into 13 to arrive at an expres-
sion for the metal ion flux in terms only of constants and the concentrations of
metal and counter ion in the aqueous solutions on the two sides of the mem-
brane. The solution is simple only when n = 1, in which case:
JMRn =
? [( [R]“,~~]*~~l+l)o - ([A]l~/[:I”K1+1) J (15)
This shows the coupling effect. Thus, there will be a positive “uphill” flux of
metal ion from the downstream to the upstream solution (i.e., in the direction
from II to 0) as long as:
(S),>(g), (16)
When the inequality is of the opposite sense, the metal ion flux is in the conven-
tional or “downhill” direction. It is also possible to determine the maximum
concentration factor, i.e., the point at which metal ion flux ceases, in terms of
the hydrogen ion concentration in the two aqueous phases:
(17)
This is, of course, identical to Equation 12 for the case of a monovalent metal
ion.
It can be easily shown that for co-transport governed by the general equa-
tion:
A+M+R ZRAM
The equivalent expressions to Equations 15 and 16 are:
iflRAMy
(18)
&AM = -
II
[(
and
-[Ml;; =-
[Al;
(IQ)
[Mlh: [A];;
where K’ is the equilibrium constant defined as:
CRAM]'
K' = (20)
[R]‘[A]“[M]”
Thus, in co-transport the gradient in A and M are in the same direction.
CHARACTERISTICS OF COUPLED TRANSPORT MEMBRANES
In this section, we will discuss the major characteristics of coupled transport

systems. The discussion will be illustrated with results obtained with supported
liquid membranes. However, the same principles apply to emulsion membranes.
We use supported liquid membrane results because the geometry of supported
membranes is well-defined and it is possible to maintain the conditions of the
feed and product solutions constant. This allows parametric studies to be per-
formed and interpreted easily. The thickness and area of emulsion membranes
are much more difficult to define, and the composition of the feed and product
solutions changes continuously. This makes the theoretical interpretation of
results obtained with emulsion membranes difficult to rationalize.
Experiments described in this section generally utilize LIX 64N@ (Henkel
Corp.), LIX 54@ (Henkel Corp.), and Kelex lOO@ (Sherex Chemical Co.) ox-
imes, used in the counter transport of copper and other metal cations by the
mechanism shown in Figure 9.6, or Alamine 336@ (Henkel Corp.), a tertiary
amine used in the co-transport of anions by the mechanism shown in Figure
9.7. These reagents are described in more detail in a later section of this chapter.
Concentration Effects
Equations 8 through 20 provide a basis for rationalizing the principal fea-
tures of coupled transport membranes. Thus, it follows from Equations 15 and
18 that coupled transport membranes are able to move metal ions from a dilute
to a concentrated solution against the metal ion concentration gradient provided
there is a sufficient gradient in the second coupled ion. A typical experimental
result demonstrating this unique feature of coupled transport is shown in Figure
9.8. The reaction is counter transport of copper driven by hydrogen ion, as de-
scribed in Equation 1. In this particular experiment, a pH difference of 1.5 units
is maintained across the membrane. The initial product solution copper concen-
tration is higher than the feed solution concentration. Nonetheless, copper dif-
fuses against its concentration gradient from the feed to the product side of the
membrane. The ratio of the counter hydrogen ions between the solutions on ei-
ther side of the membrane is about 32 to 1 which, according to the appropriate
form of Equation 12, should give a copper concentration ratio:
cd+] ‘k
-... = : 1000 (21)
(W*
ccu*+1;
In the experiment shown in Figure 9.8, this means the feed solution copper con-
centration should drop to just a few ppm, and this is the case.
A more convenient method of measuring the copper concentration factor
is to maintain the product solution at some high copper concentration factor
and allow the feed solution copper concentration to reach a steady state value.
Figure 9.9 shows the feed copper concentration in this type of experiment. In
this example, the product solution was maintained at 9.3% and 3.0% copper and
the feed solution was allowed to approach the steady state from both directions,
i.e., with an initial copper concentration higher and lower than the predicted
value for the given pH gradient. As Figure 9.9 shows, regardless of the starting
conditions, the final steady state feed copper concentration was the same. In
fact, the copper concentration factors measured this way are in reasonable agree-
ment with the predictions of Equation 21 which can be restated in another form
as:
hIg(,$$) = log ($)*= *(pi;- pH;) = 2ApH (22)

Product Solu t Ion
s
2
f 1600
!
s
i; 1000
::
s
600
0
0 1 2 3 4 6 6
Tlms ( 1 O3 mtn)
Figure 9.8: Demonstration of coupled transport. In a two-compartment cell,

copper flows from the dilute (feed) solution into the concentrated (product)
solution, driven by a gradient in hydrogen ion concentration.20 (Membrane:
Microporous Celgard 24OO/LIX 64N. Feed: pH 2.5. Product: pH 1.0).
120 -
Product 0.3%
Product 3.0%
0 1 2 3 4 5 6 1 a
Time ( lo3 mln)
Figure 9.9: Experiments to demonstrate the maximum achievable concentratior,

factor.61 (Membrane: Microporous Celgard 24OO/LIX 64N. Feed: pH 2.5, 0 or
100 ppm copper. Product: pH 1.0, 3.0% or 9.3% copper).
Concentration factors were measured as a function of the pH difference. In one

set of experiments, the downstream pH was maintained at 1.0 throughout, while
the feed pH was varied from 1.5 to 3.0. In a second set of experiments, the feed
pH was maintained at 2.5 while the downstream pH was varied from 0.5 to 2.0.
The experimental results are plotted in Figure 9.10 along with the theoretical
line obtained from Equation 22. While there is some scatter, the agreement be-
tween theory and experiment is generally good.
10,000 I 1 1
1000
100
10
0.0 0.5 1.0 1.5 2.0
APH
Figure 9.10: Maximum concentration factor vs pH difference across the mem-

brane. The line is calculated from Eq. 27.” (Membrane: Microporous Celgard
24OO/LIX 64N. Feed: 100 ppm copper. Product: 0.5 to 9 wt % copper).
Feed and Product Metal Ion Concentration Effects

A second characteristic of coupled transport membranes is that the mem-
brane flux usually increases with increasing metal concentration in the feed solu-
tion but is usually independent of the metal concentration in the product solu-
tion. This behavior follows from the flux Equations 13 and 15. In typical cou-
pled transport experiments, the concentration of the driving ion A in the prod-
uct solution is very high. For example, in coupled transport of copper, the driv-
ing ions are hydrogen ions and 100 g/Q sulfuric acid is used as the product solu-
tion. As a result, the term [MR,] h is very small compared to [MR,] c and Equa-
tion 15 reduces to:
DMRny 1
JMRn = - (23)
9” [A];I/[M];; K’ + 1
The key feature of Equation 23 is that the term [Ml i does not appear in the
flux equation. Thus, the flux is independent of the concentration of metal on
the product side of the membrane. However, the flux does depend 07, the con-
centration of metal on the feed solution side of the membrane [MI c. At low
values of [MI :, the flux will increase linearly with [MI 6, but at higher concen-
trations the flux reaches a plateau value as the term [Al z/[Ml :K’ becomes
small compared to 1. The form of this dependence is illustrated for the feed and
product solution metal ion concentration in Figure 9.11.
0 1 2 3 4 5
Product Copper Concentration (wt%)
15
10
0
0 500 1000 1500 2000
Feed Copper Concentration (PPm)
Figure 9.11: Effect of metal concentration in the feed and product solution on
flux *’ (Membrane: Microporous Celgard 2400/30% Kelex 100 in Kermac 470B.
Feed: pH 2.5. Product: 100 g/Q H,SO,).
pH and Metal Ion Effects

It follows from flux Equation 23 that the concentration of the counter hy-
drogen ion and the equilibrium coefficient K’ for a particular metal ion will af-
fect the metal ion flux. The effect of these factors can best be understood by
looking at the metal ion extraction curves vs counter ion concentration. These
are shown in Figures 9.12 and 9.13 for copper and other metals with a number
of liquid ion exchange reagents. The counter ion is hydrogen, and these reagents
extract metal ions via reactions of the type shown in Equation 1.
80
z
P
;j 80
!!
r;
W
; 40
B
0”
20
0 2 4 8 8
PH
Figure 9.12: Copper extraction vs PH by various complexing agents.
80
PH
Figure 9.13: Metal extraction curves for four metal ions by LIX 64Pl. The
aqueous phase initially contained 1,000 ppm metal as the sulfate salt.
Consider the copper extraction curves shown in Figure 9.12. These results
were obtained by extracting 0.2% copper solutions buffered to the appropriate
pH with an equal volume of organic complexing reagent. The percentage of the
copper concentrated into the organic phase was then measured. The extraction
curves for different reagents have the same form, but the pH extraction iso-
therms shift depending on the distribution coefficient for the particular agent.
At low pH, copper is not extracted because the term [H’] “*/[Cu*‘] ” in Equa-
tion 10 is large and thus the term [CUR,] ‘/[RHl ‘* is small. As the pH increases,
the term [H+l “2/[Cu2+l ” decreases and thus [CUR,] ‘/[RHl ‘* increases, produc-
ing an increase in copper metal extracted. At very high pH levels, a decrease in
copper extraction occurs because copper ions in the aqueous solution begin to
form non-extracting basic complexes with other anions in the solution.
The pH at which metal ions are extracted also depends on the distribution
coefficient for the particular metal and complexing agent. As a result, the pH
at which the metal ions are extracted varies, as shown by the results in Figure
9.13. This behavior allows one metal to be separated from another.
The metal extraction curves described above mirror the behavior of these
complexing agents when incorporated into coupled transport membranes. Thus,
when the product solution pH is maintained at a very low level, the term [MR,l b
in Equation 13 is reduced to zero and the flux is determined by the concentra-
tion of complexed metal in equilibrium with the feed solution [MR,,];. As a
result, the flux vs pH behavior follows the metal extraction curves. This is shown
in Figure 9.14, which shows the flux pH (curves) for copper with each of the
three complexing agents whose extraction curves are shown in Figure 9.12. The
magnitude of the coupled transport flux depends on the liquid viscosity and
complexed metal diffusion coefficient, but the pH dependence of the flux is
very similar to the metal ion extraction curves.
The ability of the complexing agents used in coupled transport to complex
metals at different pH levels also allows good separation to be obtained. For
example, consider the separation of copper and iron with LIX 64N. As Figure
9.13 shows, LIX 64N extracts copper at pH 1.5 to 2.0, while iron is not ex-
tracted until above pH 2.5. The separations obtained when 0.2% solutions of
copper and iron are tested with a LIX 64N membrane at various pH’s are shown
in Figure 9.15. The copper flux exceeds the iron flux by approximately IOO-fold
at a feed pH of 2.5. The copper and iron fluxes through the membrane are not
completely independent, as the results in Figure 9.16 show. Thus, a 1.0% iron
feed solution in the absence of copper has a flux of 0.15 /.rg/cm* min, but this
flux drops to 0.02 pg/cm* min when 0.2% copper is added and is restored to
the original value when the copper is removed. The phenomenon is known as
“crowding”, a term taken from the solvent extraction literature. Copper reacts
preferentially with the complexing agent, reducing the amount of reagent free
to extract iron and hence the iron flux. The phenomenon is fairly common and
is important in the separation of uranium from vanadium.
Complexing Agent Effects

Complexing agents used in coupled transport membranes are usually diluted
with a carrier solvent, typically a mixed aliphatic-aromatic hydrocarbon. We
would generally expect the amount of_ metal extracted by the complexing agent
solution to increase with increasing agent concentration, and this is usually the
80
71OVOl%
LUh.4
Kl0rmac(
I,
s
3Ovol’L Kelex@lOO \ 1700
4
in Kermacb4700
t 60
P
E
3‘0
\ to =
1
;IA
tl
Pp
8
- 2!O
-0 ,
4 6 a 10 12
Feed pti
Figure 9.14: Effect of pH on copper coupled transport flux for three different
complexing agents.” (Membrane: Celgard 2400. Feed: 0.2% copper. Product:
100 g/Q HzS04).
2.0 3.0 4.0
pH of Feed Solution
Figure 9.15: Copper and iron fluxesvsfeed PH.” (Membrane: Celgard 24OO/LIX
64N. Feed: 0.2% metal. Product: pH 1.0).
528 _Handbook of Industrial Membrane Technology
case. For example, the results in Figure 9.17 show the extraction of uranium by
a tertiary amine following the reaction shown in Equation 4. It follows that the
coupled transport flux would also increase with increasing carrier concentration.
0
0 200 400 600 800 1000
Time (mid
Figure 9.16: The effect of small concentrations of copper in the feed on iron
permeation through a coupled transport membrane.lg (Membrane: Celgard 2400/
LIX 64N. Feed: pH 2.5. Product: pH 1.0).
0
0 20 40 00 80 100
Agent Concsntratlon (VOIX)
Figure 9.17: The effect of complexing agent concentration in the organic ex-
traction phase on the amount of uranium extracted from an aqueous solution.”
(Organic Phase: Alamine 336 dissolved in Aromatic 150. Aqueous Phase: 0.2%
uranium, pH 1.0).
However, in coupled transport a more complex behavior is observed. Figure 9.18

shows a plot of the uranium flux vs the same amine in a membrane. The flux
reaches a maximum value at about 30% amine in the diluent and decreases with
increasing concentration thereafter. This appears to be a general phenomenon as,
for example, Figure 9.19 illustrates with the flux of copper through various rea-
gent-diluent solutions. The pattern of the results is similar to Figure 9.18.
0 20 40 20 20 100
Agent concentratton (VOIX)
Figure 9.18: The effect of complexing agent concentration on the coupled

transport flux of uranium.” (Membrane: Celgard 2400/Alamine 336 dissolved
in Aromatic 150. Feed: 0.2% uranium, pH 1.0. Product: pH 4.5).
0 20 40 20 20 100
Aoent Concentration (VOl%)
Figure 9.19: The effect of complexing agent concentration on the coupled

transport flux of copper. 19r2’ (Membrane: Celgard 2400/various agents dissolved
in Kermac 470B. Feed: 0.2% copper, pH 2.5. Product: 100 g/Q H,SOJ.
Two competing factors apparently caused this phenomenon: the concen-

tration gradient of the metal complex and the viscosity of the organic phase in
the membrane. As the concentration of complexing agents in the membrane is
increased, the ccncentration gradient of the metal complex increases as well. At
low agent concentration, this results in an increase in flux. However, the viscos-
ity of the solution also increases with increasing agent concentration, causing the
diffusion coefficient of the metal complex to decrease. At some point, the flux
increase due to the increasing concentration gradient is no longer sufficient to
overcome the flux decrease due to the increasing viscosity. At this agent concen-
tration, the flux reaches its maximum value and subsequently decreases with fur-
ther increase in agent concentration.
The effect of viscosity on the metal complex diffusion coefficient can be es-
timated by measuring the viscosity of the organic phase and substituting this
value into the appropriate form of the Stokes-Einstein equation:
D = kT/6nrn (24)
where k is the Boltzmann constant, T is the absolute temperature, r is the molec-

ular radius of the metal complex, and a is the viscosity of the organic phase. This
calculation has been performed for uranyl ions and the amines used to obtain
the results shown in Figure 9.17 and 9.18. To model the organic phase at the
feed interface, the viscosity of solutions that had been equilibrated with the feed
solution were measured. To model the organic phase at the product solution in-
terface, the viscosity of the uncomplexed amine ions were measured. The results
are shown in Figure 9.20. There is a large increase in viscosity with increasing
agent concentration, especially when the agent is complexed with uranium.
10.000
g 1,000
!
iz
2
i 100
6
i
‘d
a.
.= 10
B
B
>
1 b
0 20 40 00 80 100
Agent Concsntration(VOW
Figure 9.20: The effect of agent concentration in the organic phase on viscosity
for solutions containing no uranium (uncomplexed agent) and solutions pre-
viously equilibrated with aqueous uranium solution (complexed agent).” (Or-
ganic phase: Alamine 336 dissolved in Aromatic 150).
Substituting the molecular radius and the viscosity into Equation 24 gives
the diffusion coefficient for the uranium complex. Because of the wide range of
viscosities, there is a wide range of calculated diffusion coefficients. On the feed
side of the membrane, the diffusion coefficient varies from about 4 x 10”
cm*/sec for dilute solutions to 6 x lob9 cm2/sec for concentrated solutions. For
each agent concentration, then, the mean diffusion coefficient is calculated by
simply taking the arithmetic mean at the two interfaces. (Use of the arithmetic
mean diffusion coefficient is actually justified only when the concentration gra-
dient of a permeant across the membrane is linear. In coupled transport mem-
branes, the concentration gradient would vary as a result of variations in viscos-
ity across the membrane, Calculation of the true mean diffusion coefficient
would require knowledge of the concentration profile within the membrane,
which is difficult to determine.)
These calculated mean diffusion coefficients are then substituted into Equa-
tion 13, together with the measured values of the concentration of uranium
complex and the membrane porosity. The calculated flux vs agent concentration
is plotted in Figure 9.21. The measured flux is also shown for comparison. It is
clear from this figure that the calculated dependence of flux on agent concen-
tration agrees qualitatively with the measured flux dependence. However, the
calculated fluxes are as much as five times larger than the experimental values.
Various reasons have been suggested for this discrepancy, which is even larger
with some other metal complexing agents. Hindered diffusion of the metal com-
plex in the rather small pores of the support membrane is a possibility.
CalculatedFluxEared
on Stokea-Einataln
0 20 40 60 a0 100
AgentConcentration (~01%)
Figure 9.21: The effect of agent concentration on the uranium flux through a
coupled transport membrane calculated from Equations 13 and 24. The meas-
ured flux data are shown for comparison. ” (Membrane: Celgard 2400/Alamine
336 dissolved in Aromatic 150. Feed: 0.2% Uranium, pH 1.0. Product: pH 4.5).
Interfacial Reaction Rate Effects

The model for describing coupled transport developed in Equations 8 to
20 contains an implicit assumption that diffusion of the metal agent complex
is the rate-limiting step in metal transport through the liquid membrane. This
is not always the case. First, the reactions at the feed and product interfaces of
the membrane may be comparable in rate of diffusion through the membrane,
resulting in partially reaction-rate-controlled transport. In addition, the rate of
diffusion of aqueous reactants to and from the interface through the aqueous
boundary layers at the feed and product sides of the membrane may be com-
parable to the rate of diffusion through the membrane. In this case, the process
is partially boundary-layer- or concentration-polarization-controlled.
Figure 9.22A illustrates the purely diffusion-controlled process, in which
the effects of boundary layers and interfacial reaction rates are negligible. In
this case, the concentrations of the complex at the interfaces are the equilibrium
concentrations. Figure 9.2213 illustrates the partially boundary-layer-controlled
case. Here, prior to steady state, the permeant diffuses across the membrane
faster in the feed-side boundary layer and accumulation of permeant in the
product-side boundary layer. The consequence of this concentration polariza-
tion is a reduction in the net concentration gradient across the membrane, and
a reduced flux compared with the diffusion-controlled case. The last case is
that of partially reaction-rate-controlled flux, illustrated by the concentration
profile in Figure 9.22C. Here, either the permeant initially diffuses away from
the feed interface faster than it can be replenished by the interfacial reaction,
or the dissociation reaction is not fast enough to prevent accumulation of the
complex at the product interface. Again, the net result is a decrease in the con-
centration gradient compared with that in the purely diffusion-controlled case.
In all three cases, the flux is proportional to the slope of the concentration pro-
file across the liquid membrane.
The presence of interfacial reaction effects can be determined by measuring
the flux through nominally identical membranes differing only in thickness.
With diffusion control, the flux is inversely proportional to the membrane thick-
ness, and thus a plot of flux vs l/12 is a straight line passing through the origin.
This type of plot is shown in Figure 9.23 for coupled transport of copper with
three different reagents. The plot is completely linear for LIX 64N, showing
complete diffusion rate control. However, the Kelex 100 and LIX 54 curves
show some deviations, with thin membranes (high values of I/!?) showing some
interfacial effects with these reagents. An extreme example of interfacial reac-
tion rate control is coupled transport of nickel in which the flux is essentially
independent of membrane thickness unless the membrane is very thick. A plot
of nickel flux vs l/J? has the form shown in Figure 9.24.
Concentration Polarization Effects

As Figure 9.22 shows, concentration polarization due to inadequate stirring
of the aqueous solution on either side of the coupled transport membrane could
also cause concentration gradients to build up in the membrane boundary layer,
decreasing the membrane flux. In coupled transport, two permeating species are
involved and the flux of one permeant is intimately connected to the flux of
the other. Hence, concentration polarization of either species will affect the flux
of the other. For example, in coupled transport of copper with hydrogen ions,
following Equation 1, the copper flux may be limited by concentration polariza-
tion of either copper or hydrogen ions, depending on their relative concentra-
tions in the bulk solutions. At low copper concentrations, concentration-polar-
ization of copper may be rate controlling, while at high copper but low hydro-
gen ion concentration, concentration polarization of hydrogen ion may be rate
controlling.20
Feed Liquid Product

Solution Membrane Solution
I A I !
n
I ’I
I
I’
I
I
I
I
1 , IFndar y Layer
DIFFUSION CONTROLLED
-4
0
._ ! t\
I
I
I j
I
I
B I
4 I
\
BOUNDARY ‘-I BAYER CONTROLLED
4-
\ I
\
\
\
II
I
REACTION-RATE CONTROLLED
Figure 9.22: A schematic diagram showing the concentration profile of per-

meant in a liquid membrane when the orocess is (A) diffusion-controlled, (5)
boundary-layer-controlled, and (C) reaction-rate controlled. The concentration
profile in the membrane in absence of boundary-layer and reaction-rate effects
is shown for comparison.17
60 I I I
50 IOvol% LIX@54
In Kemac@47OB
2 40
E
N
5
‘D 30
t
ii 30~01% K&x@ 100
ii
g 20
2
s
10
0
0 100 200 300 400
Inverse Membrane Thickness (cm-‘)
Figure 9.23: The effect of membrane thickness on flux through a coupled trans-
port membrane.20 (Membrane: Laminated Celgard 2400/various reagents. Feed:
0.2% copper, pH 2.5. Product: 100 g/R H2S0J.
0 ,T
0.1 -
O-
0 100 200 300 400 500
Inverse Membrane Thickness (cm-‘)
Figure 9.24: The effect of membrane thickness on the coupled transport flux
of nickel through reaction rate limited membrane.73 (Membrane: Laminated
Celgard 2400/30% Kelex 100 dissolved in Kermac 4706. Feed: 0.2% nickel, pH
6.0. Product: 100 g/Q H2S04).
In general, concentration-polarization effects are much more noticeable on

the feed side of the membrane than on the product side. As discussed earlier,
the trans-membrane flux is relatively insensitive to the metal ion concentration
in the product solution but varies substantially with feed solution metal ion con-
centration, particularly with low concentration feed solutions. Concentration
polarization effects can be a serious problem when treating these solutions with
practical membrane systems. Figure 9.25 shows the effect of concentration po-
larization with a hollow fiber supported membrane system described in the fol-
lowing section. In this system, a feed solution velocity of greater than 20 cm/set
is required to overcome polarization.
oOs
* 0 10 20
Feed Flow Velocity
30
(cm/set)
40
Figure 9.25: The effect of feed solution flow velocity on coupled transport
metal flux. The product solution flow was held constant at 0.06 cm/set.”
(Membrane: Polysulfone hollow fiber/Kelex 100. Feed: 0.2% copper, pH 2.5.
Product: 2% copper, 100 g/Q H2S04).
In cases when the H+ or OH- is the diffusing counter ion to metal transport,
buffer ions in the surrounding aqueous solution can effect the extent of concen-
tration polarization. The mechanism by which a buffer can reduce the concen-
tration of hydrogen ions in the boundary layer is shown in Figure 9.26. In the
bulk solution, the pH is high and the acid of the buffer is in its ionized form,
denoted B-. Near the membrane surface, the pH is lower, due to concentration
polarization of hydrogen ion, and the acid is in its protonated form, BH. Thus,
B- diffuses toward the membrane surface down its concentration gradient and
BH diffuses away from the membrane surface down its concentration gradient.
In this way, B- acts as a shuttle, picking up hydrogen ions near the membrane
surface, transporting them out of the boundary layer, and depositing them in
the bulk solution.
BH - B-‘ Ii+
Membrane
Figure 9.26: Schematic representation of the buffer cycle and pH gradient

within the boundary layer of a coupled transport membrane.”
For this mechanism to be effective, the pH at the membrane surface

must be maintained above a certain value. Below this pH, copper extraction
by the ion exchange agent will be reduced and the copper flux will decrease.
This pH is about 1.5 for Kelex 100, as determined from flux vs pH data shown
in Figure 9.12. Thus, in experiments where the bulk feed solution has a pH of
2.5, the most effective buffer would be one in which the acid is ionized at a
pH of 2.5 and protonated at a pH of 1.5, i.e., it should have a pk, of about 2.
This pk, will maximize both the BH concentration gradient from the membrane
surface to the bulk solution and the B- concentration gradient from the bulk
solution to the membrane surface. A weak acid with a pk, of 4, for example,
would be essentially fully protonated both in the bulk solution at pH 2.5 and at
the membrane surface at pH 1.5. Thus, there would be no concentration grad-
ient of either B- or BH. A strong acid, with a pk, of 0, for example, would be
ionized in both the bulk solution and at the membrane surface. Again, there
would be no concentration gradient of either B- or BH.
The results of experiments to test this mechanism by determing the ability

of buffers with different pka values to reduce concentration polarization effects
are shown in Figure 9.27. Low stirring rates (20 rpm) were employed to maxi-
mize the concentration polarization effect. The feed solutions contained 2,000
ppm copper and were buffered at pH 2.5, and the product solutions were 100
g/Q sulfuric acid. (The buffer concentration refers to the initial amount of acid
added to the solution. For example, a 1 molar acetate buffer would be a 1 molar
solution of acetic acid to which sufficient NaOH was added to raise the pH to
2.5.)
As expected, the copper flux is most affected by buffer acids with a pka
near 2.0. Thus, when a feed solution containing 0.05 molar formate buffer (pka
3.75) is used, the copper flux is the same as when no buffer is present. When a
feed solution containing 0.05 molar dichloroacetic acid buffer (pk, 1.48) is
used, the flux is more than 4 pg/cm’-min higher than that for unbuffered SOIU-
tions.
16
0
0 0.1 1.0
Buffsr con~antratlon hoIN
Figure 9.27: The effect of different buffers on coupled transport metal flux.*’
(Membrane: Celgard 2400/Kelex 100. Feed: 0.2% copper, pH 2.5, 0.05 molar
buffer. Product: 100 g/R H2S04).
COUPLED TRANSPORT COMPLEXING AGENTS
In the examples given above to illustrate coupled transport, various oxime

carriers for copper and other cations and tertiary amines as carriers for uranyl
anions were used. However, a large number of complexing agents have been de-
scribed in the literature; some of them are listed in Table 9.1.
Table 9.1: Coupled Transport Complexing Agents.
References
Trade Name Complexed Ion
Chemical Formula Emulsion Membranes
and Manufacturers Supported Membranes
LIX 633 CL?+, NiEt,

Henkel Corp. co2+
Minneapolis, MN
LIX 64N?a mixture of Cu2+, Ni2+ 15, 18-20, 34. 43 27, 34, 37
a few percent of LIX 61, 62, 63
6@in LIX 65@
Henkel Corp. "XU*
LIX 70?a mixture of cu2+ 64

LIX 63Gnd this
oxime:
Henkel Corp.
Table 9.1: (continued)
References
Trade Name
Chemical Formula Complexed Ion
and Manufacturers Supported Membranes Emulsion Membranes
I I I
Kelex lo@ cu2+, co2+,
Sherex Chemical Ni2+ 60-62 64
Company
Dublin, OH
SME 5298 cu2+ 43

-
Shell Chemical Q \ / -*
Company s
Houston, TX G
z
Alamine 3368 UO2(So4)=- 16, 17, 61, 62, 70 P
%- -i
Henkel Corp. Cr2072- 65, 66 2
r...csnu..~.*" 3
Gl
Acorga P17. P50, cu2+ 27 5
P5100, P5300@
5
Acorga, Ltd. 3
London, UK ,R 4
a
$
Table 9.1: (continued)
I References
Trade Name
Chemical Formula Complexed ion Emulsion Membranes
and Manufacturers Supported Membranes
I
Amberlite LA-2@ Cl- 4-6
Rohm A Haas ";c*,r..rpc
Philadelphia, PA -'W-c.+
Monesin-cholanic Li+. K+, cs+, 30-32

acid
Nat
Various
Cephalin 1, Nat, Kt 71, 72

Various +.z."_+$S
*
Tributyl- NO3' 7-12

phosphate 0-a
Various II
Triocytylmine
Various
z- 67
Z?-
Crown ethers See Table 9.2 Alkali metal 30, 33, 48-51
cations
Various
Table 9.2: Some Typical Macrocyclic Crown Ethers Used in Coupled Transport.so
r-0
0
OLo<
0
n = 1: 16.Crown-6 ( 16C6)
n=2: 21.Crown-7 (2lC7)
c J3 a:>bJ
Dicydohexeno-1 B-Crown-6
(DC1 6C6)
0
Benzo-16-Crown-6
(I31 ecsj
Cikelopyr~dmo~16Crown-6 n= 1: 4Methoxydiketo- x-H: 1.10~Diaza-16.Crown-6

(OKP18C6) pyridino. 16.Crown-6 12.21
(4.MoOKPl6C6) x-C&,,: N.N-Oldecyl-1.
n=2: 4-Melhoxydlketo- 1O.Omza. 16.Crown-6
pyridino.2 1-Crown.7 (2.2DDI
Crypland (2 2 11 Crypland (2 2.2)
Dibenzo-24.Crown-6
(062403)
Cryptend [2.2.20( Cryp(and (2.2.261
MEMBRANE SYSTEM APPLICATIONS AND DESIGN
Applications
The principal applications of coupled transport have been the separation

and concentration of metals from hydrometallurgical feeds or industrial effluent
streams. These streams and the membranes used are briefly described below. In
general, the application of a coupled transport process to mining operations in-
volves the installation of a very large plant, and mine operators are reluctant to
risk this type of investment in as yet unproven processes. Thus, the first com-
mercial applications of coupled transport are likely to be smaller plants installed
in pollution control applications.
Copper Recovery. One of the most extensively studied applications of cou-
pled transport is in the recovery of copper from in situ dump and heap leach
streams. These streams are produced by the extractions of low grade copper ores
with dilute sulfuric acid. Typically, the leach stream contains 1,000 to 5,000
ppm copper and various amounts of other metal ions, principally iron, Cur-
rently, copper is removed from these streams by cementation or solvent ex-
traction. A scheme for recovering the copper by coupled transport is shown in
Figure 9.28. The 1,000 to 5,000 ppm copper solution from the dump forms the
feed solution, while concentrated HzS04 from the electrowinning operation
forms the product solution. Copper from the feed solution permeates the mem-
brane, producing a raffinate solution containing 100 to 200 ppm copper. This
solution is returned to the dump. The product solution, which contains 3 to 5%
copper, is sent to the electrowinning tankhouse.
A number of papers have been published describing the application of cou-
pled transport to this application with supported15~1s~63and emulsion27~37~41
membranes. The economics of the process appear to be promising, but the cop-
per mining industry is currently very depressed, making installation of a coupled
transport plant in the near future unlikely.
Ore Leaching
Separation Product
and Recovery
Concentration EIectrowinning)
Figure 9.28: Schematic of copper recovery by coupled transport from dump

leach streams.
Cobalt and Nickel Recovery. Cobalt and nickel are relatively valuable metals
often found in complex ores such as laterites or deep sea nodules. The metals
can only be extracted from these ores by hydrometallurgy. A proposed recovery
scheme based on coupled transport is shown in Figure 9.29. The first membrane
contains LIX 54, which produces a nickel and copper concentrate and a cobalt
raffinate stream. The concentrate stream is then passed to a second Kelex 100
membrane, which produces a copper and nickel stream. The cobalt III raffinate
stream is neutralized and reduced to cobalt II, which can then be concentrated
by a LIX 51 membrane.
Membrane II
Copper
Metal
Nickel
Metal
Dilute
co+++ Membrane III
XI-5’1 m
I
co++- Cobalt
Rec. Metal
Figure 9.29: An integrated coupled transport processing scheme using mem-

branes containing LIX 54, Kelex 100 and Xl-51.62
Uranium Recovery. The application of coupled transport to uranium re-

covery is illustrated in Figure 9.30. Typically, a tertiary amine such as Alamine
336 is used as the complexing agent and transports uranyl sulfate ions to the
concentrate stream by co-transport.65 These uranium feed streams are obtained
by acid leaching of uranium ores or as by-product streams from the extraction
of other metals, for example, copper. The major impurities are vanadium, moly-
bdenum and iron. Vanadium and iron are present as cations in acid feed solu-
tions and are not transported by amine complexing agents. Molybdenum and
uranium are both present as anions and are both concentrated as shown in Fig-
ure 9.30. Separation of uranium from molybdenum is achieved by selectively
precipitating the uranium with hydrogen peroxide to form yellow cake UO+
The economics of this scheme have been tested in a 300 ft’ supported membrane
pilot plant installed at a Wyoming uranium mill, but the results are not yet avail-
able.
Electroplating Rinse Waters. The general way in which coupled transport
might be applied to electroplating rinse waters is shown in Figure 9.31 for a
chrome-plating operation.66 A typical chrome-plating bath contains 25 wt %
chromium. The plated parts constantly drag out this solution into the rinse
baths. Normally the rinse water flow rate is set at 50 times this drag-out rate,
producing a CrOa concentration of 5,000 ppm in the first rinse bath. It is this
solution that is fed to the coupled transport concentrator. An amine-containing

membrane can be used, and, if the product side of the membrane is maintained
at an alkaline pH with NaOH, then chromate anions are transported across the
membrane by co-transport via the reaction:
2R3N + 2H+ + Cr04 _-( R3NH)2 Cr04 (25)
The raffinate from the process is thus a chromate-depleted solution, which can
be recycled to the rinse baths, and a concentrated sodium chromate solution,
from which the chromate can be recovered. Similar schemes have been proposed
for copper, zinc, mercury and nickel rinse waters.25~26~66-70
Feed Solution (U,V,Mo,Fe)
Na 2CO3 Raffinate W,Fe)
Coupled Transport
Concentrator
Product (U.MO)
Yellow Cake (VI Filtrate (MO)
Concentration (ppm)
U V MO Fe
Feed Solution 2000 100 100 1140
Rafflnate 13 100 3 1140
Product 30,000 0.1 1500 0.1
Figure 9.30: A processing scheme and results for the coupled transport separa-
tion of a synthetic uranium mine leach stream. Metals were added to the feed
solution as UOzSO4, NaV03, Na2M,04, FeS04, and Fe2(S04)3.66
Water
Rhao SoMkm
Waste Streem
a
Draoout
c---------
Dragout
r-----l I
I 1 :
Rinse Sokrtion Reusable Water
Waste Stream
,
1
Chromate
b Concentrate
Figure 9.31: Schematic of a typical chromium plating operation (a) without

and (b) with coupled transport treatment.67
Copper Etchant Baths. One of the most promising applications of coupled

transport is the renovation of circuit board etchant solutions that contain cop-
per.61 The printed circuit board industry produces more than ten million gal-
ions/year of spent etchant solutions in which copper, other salts and etchant
chemicals are concentrated. Coupled transport permits continuous on-site re-
moval of copper from the etchant solutions and simultaneous regeneration of
the etchant solution. This represents a considerable savings in the costs of manu-
facturing circuit boards.
The use of coupled transport to continuously renovate the etchant and re-
cover copper is shown schematically in Figure 9.32. The relevant chemical reac-
tions are shown. Copper on the circuit board is dissolved by oxidation with at-
mospheric oxygen and complexation with ammonia. The etchant, containing the
copper complex, is pumped to the coupled transport processor, where copper is
extracted into the membrane while ammonia remains in the aqueous phase. (The
ammonia is not extracted into the membrane because bidentate complexing
agents are used. Two of these complexing agents occupy all four coordination
sites on the copper, thereby displacing ammonia from the copper-ammonia com-
plex.) For each mol of copper removed, two mols of water are generated. Half
the excess water is consumed in the etching reaction. Thus, except for a slight
dilution, the etching solution is restored to its original condition. This dilution
would not present any problems in the etching process because elevated tem-
peratures are used, and water must be continuously added to compensate for
evaporation.
On the product side of the membrane, concentrated sulfuric acid (100 g/Q)
is employed. Thus, copper sulfate is formed. Copper could be recovered by pre-
cipitating the copper sulfate or by electrodepositing the copper. In the latter
case, the overall oxidation-reduction reaction,
2H2g + 2Cu2+~2Cu + 4H+ + 92, (26)
would regenerate the hydrogen ions lost in the coupled transport process. If
copper sulfate precipitation were used, acid would be continuously metered
into the product solution to replace hydrogen ions lost in the coupled transport
process.
Copper recovery from etchant baths is a particularly promising application
of coupled transport because the high copper concentrations in the etchant so-
lutions result in high fluxes. A relatively small unit is, therefore, able to process
a large volume of solution.
Product Solutlon
Feed Solution
1 I
1 1
I v I _ I I
I
Etching Coupled Copper

Procesr Transport Recovery
Procerror
I
I I
I‘ I
a
Etching Procore Feed Solution Liquid Product Solution

Membrane
cu++
4 NH3 + tip
2H++ SO,--
1
H20 b
Figure 9.32: Schematic diagram of (a) coupled transport renovation of circuit
etchants and (b) relevant chemical reactions.62
Phenol and Ammonia Recovery. Finally, the closely related passive and fa-
cilitated-transport processes for phenol and ammonia recovery should be men-
tioned. In these processes, dilute phenol or ammonia feed solutions are con-
tacted with a liquid membrane in which they are freely soluble. They dissolve
in the membrane, diffusing to the product side where they are removed by
neutralizing with a base (in the case of phenol) or an acid (in the case of ammo-
nia). Although the transport mechanism does not involve a carrier and these are,
therefore, passive transport processes, the actual process is quite similar, and
Li et a12’ published the details of these separations using emulsion membrane
technique.
Supported Liquid Membranes Process Design

In supported liquid membranes, a microporous support is impregnated with
the liquid complexing agent phase and used to separate the feed and product so-
lutions. In any large process, this support membrane must be mounted in some
form of high surface area module, and the way in which the membrane is modu-
lized strongly influences both the operation and economics of the process. One
requirement of this process, in contrast to other membrane processes such as
reverse osmosis and ultrafiltration, is that the fluid on both sides of the mem-
brane must be circulated to avoid concentration polarization effects. However,
as described earlier, concentration polarization is much more significant on the
feed side of the membrane. One possible design would be a simple plate-and-
frame module. However, this is unlikely to be the optimum membrane configu-
ration for large-scale plants because of the relatively high cost per unit area. A
potentially more economical configuration uses hollow fiber membranes in a
tube-and-shell configuration and almost all of the available process data has in-
volved this type of module.
The hollow fibers used in these module configurations are usually made
from polysulfone by a solution spinning process.1912’Polysulfone is used because
of its excellent chemical resistance to acids, bases, and a wide variety of organic
compounds. However, other materials have been used, such as polyvinylidene
fluoride. A scanning electron micrograph of a cross-section of a representative
fiber is shown in Figure 9.33. Permeation occurs not only through the finger-
like voids visible in the scanning electron micrograph, but also through much
smaller pores, which are not visible at this magnification. Fibers such as these
are incorporated into modules of the type illustrated in Figure 9.34. In opera-
tion, the organic complexing agent is incorporated into the walls of the fiber,
the product solution flows on the outside of the fiber, and the feed solution
flows down the fiber lumen.
Large-scale processes require a number of modules to remove most of the
metal from a continuous feed stream. In general, a multi-stage system operat-
ing in a feed and bleed mode is the most efficient design. A schematic represen-
tation of a three-stage system is presented in Figure 9.35. A fixed feed volume
is recirculated through each module, as indicated by the bold lines in the figure.
Feed solution is continuously introduced into the recirculating volume of the
first stage and is bled off at the same rate. This “bleed” from the first stage con-
stitutes the feed for the second. The bleed from the second stage constitutes
the feed for the third. In operation, the concentration of metal in the feed so-
lution decreases as it flows from stage 1 to stage 3, with the final raffinate con-
centration depending on the feed-and-bleed flow rate. The productsolution flows
in series through the stages. The advantage of this multi-stage design over a single
stage system is that only the final stage operates on feed solution depleted of
metal.
Liquid membranes supported by hollow fibers are relatively easy to make
and operate and the membrane fluxes can be quite high. However, membrane
stability is a problem. The variation in coupled transport flux during long-term
tests is shown in Figure 9.36. As these results show, actual behavior varies quite
widely, depending on the membrane and the complexing agent. However, each
curve is quite reproducible.
The detailed mechanism for this flux instability is unknown but it appears
to be related to loss of the organic complexing agent phase from the support
membrane. Thus, it is possible to restore membrane fluxes to their original val-
ues by reloading the membrane with fresh complexing agent. Figure 9.37 shows
this result with polysulfone hollow fiber membranes loaded with Lelex 100, an
oxime complexing agent for copper-coupled transport.
200JJm
t .
50~m
I I
Figure 9.33: Scanning electron micrographs of a polysulfone hollow fiber as a

coupled transport support membrane.lg
Product Solution
Feed Soluti R~fflnrte

Solution
Hollow
Fiber
Module
Pfeerure
OsilpS3
Figure 9.34: Schematic of a hollow fiber module and test unit.*‘j6*
Stage 1 Stage 2 Stage 3
Concentrated
Figure 9.35: Schematic of a three-staged feed and bleed hollow fiber coupled
transport concentrator.*’
I I I 1 --
20
% 15
‘0
t
a
Y 10
;
8
8
5
I I I I
50 75 100 125 0
Time (days)
Figure 9.36: Copper flux as a function of time for coupled transport mem-
branes containing different complexing agents. The arrow indicates the time of
failure of the membrane.‘j (Membrane: Celgard 2400/variouscomplexing agents.
Feed: 0.2% copper, pH 2.5. Product: 100 g/Q H,SO,).
Module Reloeded
I I
0 10 20 30 40
Tlme (days)
Figure 9.37: The effect of replenishing a hollow fiber coupled transport module
with fresh complexing agent. 63 (Membrane: Polysulfone, hollow fiber/Kelex
100. Feed: 0.2% copper, pH 2.5. Product: 2% copper, 100 g/Q H2S04).
A model that is consistent with the observed results is spontaneous emulsifi-

cation of the liquid organic phase, as shown in Figure 9.38. Initially, the emul-
sion phase formed at the surface of the liquid membrane hinders the diffusion of
metal ions through the aqueous phase to the membrane surface. This would be
particularly apparent on the feed side of the membrane and would explain the
initial drop in flux often observed. As the emulsification process proceeds, some
of the droplets escape to the aqueous solutions and the emulsion layer and the
remaining liquid membrane layer gradually becomes thinner. This results in a
gradual inverse in the membrane flux until at some point the liquid phase dis-
appears completely and the aqueous feed and product solutions can mix. The
process is illustrated in Figure 9.39. The actual form of any particular flux-time
curve depends a great deal on the membrane material, its pore size and geom-
etry, and the viscosity and surface tension of the organic liquid phase.
Intact Membrane Partially Degraded Fully Degraded

Membrane Membrane
Agent-Filled he
lncreaeing Membrane Age
t
Figure 9.38: A schematic of membrane degradation by formation of an inter-

facial emulsion.63
Membrane Fallure
Lo88 Of Emulsified
Complexing Agent
TIME
Figure 9.39: Diagram illustrating spontaneous emulsification processes in a

liquid membrane and the effect of the process on coupled transport flux.
Emulsion Liquid Membranes Process Design

Figure 9.40 is a schematic illustration of an emulsion liquid membrane proc-
ess. There are four main operations in this process. First, fresh product solu-
tion is emulsified in the liquid organic membrane phase. This water/oil (w/o)
emulsion then feeds to a large mixer vessel, where it is again emulsified forming
a water/oil/water emulsion (w/o/w). Metal ions contained in the feed solution
permeate by coupled transport through the walls of the emulsion to the product
solution. A single mixer, as shown in Figure 9.40, or a series of mixers may be
used to extract the feed metal completely. The mixture then passes to a settler
tank where the oil droplets separate the metal-depleted raffinate solution. The
emulsion concentrate then passes to a deemulsifier where the organic and con-
centrated product solutions are separated. The regenerated organic solution is
then recycled to the first emulsifier stage.25~69
Concentrated
Product Sohrtiin
Fresh Product Solution
11
De-emulsifier
Figure 9.40: Schematic flow scheme of a liquid emulsion membrane process.
The optimum operating conditions for this type of process vary a great deal.
The first w/o emulsion typically consists of an approximately 50/50 mixture,
which is then mixed with the aqueous feed solution phase at a ratio of 1 part
emulsion phase to 5 to 20 parts feed solution phase. A typical extraction curve
is shown in Figure 9.41 for copper using SME 529, an oxime complexing agent
from Shell.43 The extraction rate generally follows a first order expression,
which can be expressed as:
dC = kC
X (27)
where k is the first order rate constant. This expression rearranges to:
In +. = kt (28)
0
where C is the concentration of copper in the feed solution at time t and Co is
the initial copper concentration. A plot of the results following Equation 28 is
shown in Figure 9.42.
Time (min)
Figure 9.41: Variation of copper concentration in the raffinate during extrac-

tion with a liquid emulsion-membrane compiexing agent system.43
5oa
200
F 100
9
E
._
t! 50
E
ifi
0g 20
ii
aQ 10
s
1 1
4 8 12 18 1
Time (min)
Figure 9.42: Variation of copper concentration with time showing a pseudo
first order rate process for the results shown in Figure 9.41.43
The slope of the curve in Figure 9.42 is proportional to the loading of com-
plexing agent in the organic phase and the rate of agitation of the mixer vessel.
Typical results illustrating this behavior for copper with LIX 64N as the com-
plexing agent are shown in Figures 9.43 and 9.44.
300
‘ii
8
g 200
%
b I% LlX@64N in Kerosene
S
t?
8
B
:: 100
s LIX’64N in Kerosene
0
.O 20 40 80 60 100
Time (min)
Figure 9.43: Copper extraction by means of a liquid emulsion membrane proc-

ess 37 Experimental conditions are as follows. Feed: 200 ml, pH 2.0, 300 ppm
cu;+. Membrane: 15 ml 10% LIX 64N in Kerosene, 3% Span 80. Stripping solu-
tion: 15 ml HzS04 solution. Stirrer speed: 500 rpm.
250 rpm
J
ia; _-
\
750 rpm 500 rpm

\ ‘(
Y
20 40 60 60 100
Time(min)
Figure 9.44: Copper removal at different agitation speeds for a liquid emulsion
process.37 Experimental conditions are as follows. Feed: 200 ml, pH 2.0, 300
ppm copper. Membrane: 15 ml 10% LIX 64N in Kerosene, 3% Span 80. Strip
ping solution: 15 ml HzH04 solution.
Figures 9.43 and 9.44 also illustrate one of the major problems of emul-
sion membrane systems, i.e., degradation of the emulsion on prolonged contact
with the feed solution and high-speed mixing of the product and feed solutions.
Prolonged stirring of the emulsion with the feed solution causes the copper con-
centration in the feed solution to rise as some of the emulsion droplets break.
Careful tailoring of the stirring rate and surfactant composition is required in
order to minimize premature breaking of the emulsion.
Although emulsion degradation must be avoided in the mixer and settler
tanks, complete and rapid breaking is required in the deemulsifer where the
product solution is separated from the organic complexing agent. Currently,
electrostatic coalescers appear to be the best method of breaking these emul-
sions. Nonetheless, some loss of the organic phase occurs with the feed solution
raffinate.
THE FUTURE
Although coupled transport has been known for many years, it has only
oeen under active development since the mid-1960’s. In the laboratory, the
process can perform a number of useful separations, but the process has yet to
prove itself industrially. The principal problems are related to the stability of
the liquid membranes, leading to poor economics either because of loss of rea-
gents or replacement of the support membranes. If these problems can be
solved, the process may find industrially important applications.
REFERENCES
1. Babcock, W.C., Kelly, D.J., LaChapelle, E.D., Smith, K.L. and Baker, R.W.,
“Coupled Transport Membranes in Hydrometallurgy,” The Proceedings
of Hydrometallurgy. Society of Chemical Industry, Manchester, England
(1981).
2. Pfeffer, W.,Ges. Wk., 16, 185 (1890).
3. Osterhout, W.J.V., Proc. Nat. Acad. Sci., 21, 125 (1935).
4. Sollner, K. and Shean, G.M., J. Am. Chem. Sot., 86, 1901 (1964).
5. Shean, G.M. and Sollner, K., “Carrier Mechanisms in the Movement of Ions
Across Porous and Liquid Ion Exchanger Membranes,” Ann. N. Y. Acad.
Sci., 137, 759 (1966).
6. Sollner, K. and Shean, G., Protoplasma, 63:1-3 (1967).
7. Bloch, R., Kedem, 0. and Vofsi, D., Nature, 199, 802 (1963).
8. Bloch, R., Finkelstein, A., Kedem, 0. and Vofsi, D., lnd. fng. Chem. Proc.
Des. Dev., 6:2 231 (1967).
9. Bloch, R., “Hydrometallurgical Separations by Solvent Membranes” in
Proceedings of Membrane Science and Technology, Columbus Labora-
tories of Battelle Memorial Institute, Columbus, Ohio, 1969, Plenum
Press (1970).
10. Vofsi, D., Kedem, O., Bloch, R. and Marian, S., J. Inorg. Nucl. Chem., 31,
2631 (1969).
11. Jagur-Grodzinski, J., Marian, S. and Vofsi, D., Sep. Sci., 8, 33 (1973).
12. Vofsi, D. and Jagur-Grodzinski, J., Naturwissenschaften, 60, 1 (1973).
13. Moody, G.J. and Thomas, J.D.R., “Poly(vinylchloride) Matrix Membrane

Ion-Selective Electrodes” in Ion-Selective Electrode Methodology, A.K.
Covington (Ed.), CRC Press, Boca Raton, Florida (1979).
14. Miyauchi, T., U.S. Patent 4,051,230 (1977).
15. Baker, R.W., Tuttle, M.E., Kelly, D.J. and Lonsdale, H.K., J. Membr. Sci.,
2, 213 (1977).
16. Babcock, W.C., Baker, R.W., LaChapell, E.D. and Smith, K.L., J. Membr,
Sci., 7, 71 (1980).
17. Babcock, W.C., Baker, R.W., LaChapell, E.D. and Smith, K.L., J. Membr.
Sci., 7, 89 (1980).
18. Largman, T. and Sifniades, S., Hydrometall., 3:2, 153 (1978).
19. Babcock, W.C., Baker, R.W., Kelly, D.J., Lonsdale, H.K., Ray, R.J. and
Tuttle, M.E., Coupled Transport Membranes for Metal Separations,
Final Report, Phases I and Il. U.S. Bureau of Mines Technical Report.
NTIS No. PB80 11 430 (May), Springfield, Virginia (1977).
20. Babcock, W.C., Baker, R.W., Kelly, D.J. and Lonsdale, H.K., Coupled Trans-
port Membranes for Metal Separation, Final Report, Phase ill. U.S.
Bureau of Mines Technical Report. NTIS No. PB293029 (May), Spring-
field, Virginia (1978).
21. Li, N.N., U.S. Patent.3,4,0,794 (1968).
22. Li, N.N., lnd. Eng. Chem. Proc. Des. Dev.. IO:2 (1971).
23. Li, N.N., J, Membr. Sci., 3, 265 (1978).
24. Li, N.N.,AIChEJ., 17:2 (1971).
25. Li, N.N. and Shrier, A.L., “Liquid Membrane Water Treating” in Recent
Advances in Separation Science, Vol. I, N.N. Li (Ed.), CRC Press, Boca
Raton, Florida (1972).
26. Frankenfeld, J.W. and Li, N.N., “Waste Water Treatment by Liquid Ion Ex-
change in Liquid Membrane Systems” in Recent Developments in Sepa-
ration Science, Vol. Ill-Part B, N.N. Li (Ed.), CRC Press, Boca Raton,
Florida (1977).
27. Li, N.N., Cahn, R.P., Naden, D. and Lai, R.W.M., Hydrometall., 9. 277
(1983).
28. Matulevicius, E.S. and Li, N.N., Sep. and Pur. Methods, 4:l. 73 (1975).
29. Hayworth, H.C., Chemtech 11:6, 342 (1981).
30. Cussler, E.L., AIChEJ., 17:6, 1300 (1971).
31. Reusch, C.F. and Cussler, E.L., AIChE J,, 19:4, 736 (1973).
32. Schiffer, D.K., Choy, E.M., Evans, D.F. and Cussler, E.L., Chem. Eng. Prog.
Sym., 70: 144, 150 (1974).
33. Caracciolo, F., Cussler, E.L., and Evans, D.F., AlChE J., 21 :I, 160 (1975).
34. Lee, Kyung-Hee, Evans, D.F. and Cussler, E.L., AIChE J., 24:5, 860 (1978).
35. Volkel, W., Wandrey, C. and Schurgerl, K., Sep. Sci., 12:4, 425 (1977).
36. Halwachs, W. and Schurgerl, K., lntl. Chem. Eng., 20:4, 519 (1980).
37. Volkel, W., Halwachs, W. and Schurgerl, K., J. Membr. Sci., 6, 19 (1980).
38. Volkel, W., Poppe, W., Halwachs, W. and Schurgerl, K., J. Membr. Sci., 11,
333 (1982).
39. Marr, R. and Kopp, A., Chem. Ing. Tech., 52:5, 399 (1980).
40. Bouvier, A., Gaubinger, W. and Marr, R., CA V, II:6 (1982).
41. Stelmaszek, J.,J. Membr. Sci., 2, 197 (1977).
42. Gladek, L., Stelmaszek, J. and Szust, J., J. Membr. Sci., 12, 153 (1982).
43. Martin, T.P. and Davies, G.A., Hydrometall., 2, 315 (1976/1977).
44. Chiarizia, R., Castagnola, A., Danesi, P.R. and Horwitz, E.P,, J. Membr.
Sci., 14, 1 (1983).
45. Danesi, P.R., J. Membr. Sci., 20, 231 (1984).
46. Folkner, C.A. and Noble, R.D., J. Membr. Sci., 12, 289 (1983).
47. Bunge, A.L. and Noble, R.D., J. Membr. Sci., 21, 55 (1984).
48. Lamb, J.D., Christensen, J.J., Izatt, S.R., Bedke, K., Astin, M.S. and Izatt,
R.M., J. Am. Gem. Sot., 102:10, 3399 (1980).
49. Lamb, J.D., Izatt, R.M., Garrick, D.G., Bradshaw, J.S. and Christensen, J.J.,
J. Membr. Sci., 9, 83 ( 1981).
50. Lamb, J.D., Brown, P.R., Christensen, J.J., Bradshaw, J.S., Garrick, D.G.
and Izatt, R.M., J. Membr. Sci., 13, 89 (1983).
51. Izatt, R.M., Haws, R.M., Lamb, J.D., Dearden, D.V., Brown, P.R., McBride,
D.W., Jr. and Christensen, J.J., J. Membr. Sci., 20, 273 (1984).
52. Way, J.D., Noble, R.D., Flynn, T.M. and Sloan, E.D., J. Membr. Sci., 12,
239 (1982).
53. Pearson, D., “Supported Liquid Membranes for Metal Extraction from Di-
lute Solutions ” in /on Exchange Membranes, D.S. Flett (Ed..), Ellis
Horwood Limited, Chichester, England (1983).
54. Chan, C.C. and Lee, C.J., J. Membr. Sci., 20, 1 (1984).
55. Cussler, E.L., Multicomponent Diffusion, Elsevier Publishing Co., Amster-
dam (1976).
56. Smith, K.A., Meldon, J.H. and Colton, C.K.,A/ChEJ., 19:1, 102 (1973).
57. Schultz, J.S., Goddard, J.D., Suchdeo, S.R.,AIChEJ., 20:3, 417 (1974).
58. Goddard, J.D., Schultz, J.S. and Suchdeo, S.R.,AIChEJ., 20:4 (1974).
59. Jain, R. and Schultz, J.S., J. Membr. Sci., 11, 79 (1982).
60. Babcock, W.C., Baker, R.W., Kelly, D.J., Kleiber, J.C. and Lonsdale, H.K.,
Coupled Transport Systems for Control of Heavy Metal Pollutants, U.S.
Environmental Protection Agency Technical Report. NTIS No. PB80-
103542, Springfield, Virginia (1979).
61. Babcock, W.C., Baker, R.W., Brooke, J.W., Kelly, D.J., LaChapelle, E.D.
and Lonsdale, H.K., Coupled Transport Membranes for Metal Recovery,
Final Report, Phase II. Technical Report, National Science Foundation
Report. NTIS No. PB81-179947, Springfield, Virginia (1980).
62. Babcock, W.C., Baker, R.W., Kelly, D.J., Kleiber, J.C. and Lonsdale, H.K.,
Coupled Transport Membranes for Metal Separations, Final Report,
Phase IV. U.S. Bureau of Mines Technical Report. NTIS No. PBSl-
214843, Springfield, Virginia.(l979).
63. Komasawa, I., Otake, T. and Yamashita, T., Ind. Eng. Chem. Fundam., 22,
127 (1983).
64. Strzelbicki, J. and Charewicz, W., Hydrometall., 5, 243 (1980).
65. Babcock, W.C., Baker, R.W., Kelly, D.J. and LaChapelle, E.D., “Coupled
Transport Membranes for Uranium Recovery” in Proceedings of ISEC’80,
Liege, Belgium (September 1980).
66. Babcock, W.C., Baker, R.W., Conrod, M.G. and Smith, K.L., “Coupled
Transport Membranes for Removal of Chromium from Electroplating
Rinse Solutions“ in Chemistry in Water Reuse, Ann Arbor Science: Ann
Arbor, Ml (1981).
67. Hochhauser, A.M. and Cussler, E.L., “Concentrating Chromium with Liq-
uid Surfactant Membranes” in Adsorption and Ion Exchange, AIDhE
Symp. Series, 71:152, 136 (1976).
68. Fuller, E.J. and Li, N.N.,J. Membr. Sci., 18, 251 (1984).
69. Halton, T.A., Lightfoot, E.N., Cahn, R.P. and Li, N.N., lnd. Eng. Chem.
Fundam., 22, 27 ( 1983).
70. Boyadzhiev, L. and Bezenshek, E., J. Membr. Sci., 14, 13 (1983).
71. Rosano, H.L., Schulmann, J.H. and Weisbuch, J.B., Ann. N. Y. Acad. Sci.,
92,457 (1961).
72. Rosano, H.L., Duby. P. and Schulmann, J.H., J. Phys. Chem., 65, 1704
(1961).
73. Babcock, W.C., Baker, R.W., Conrod, M.G., Lonsdale, H.K. and Smith, K.L.,
“Industrial Water Reuse with Coupled Transport Membranes,” Office of
Water Research and Technology Technical Report. PB No. SO-206014
(1979).
10
The Separation of Gases by Membranes
A. Keith Fritzsche and James E. Kurz
INTRODUCTION
The study of gas transport in membranes has been actively pursued for over
100 years. This extensive research resulted in the development of good theories
on single gas transport in polymers and other membranes. The practical use of
membranes to separate gas mixtures is, however, much more recent. One well-
known application has been the separation of uranium isotopes for nuclear
weapon production. With few exceptions, no new, large scale applications were
introduced until the late 1970’s when polymer membranes were developed of
sufficient permeability and selectivity to enable their economical industrial use.
Since this development is so recent, gas separations by membranes are still less
well-known and their use less widespread than other membrane applications
such as reverse osmosis, ultrafiltration and microfiltration. In excellent reviews
on gas transport in polymers as recent as 1983, no mention was made of the im-
portant developments of the last few years. For this reason, this chapter will
concentrate on the more recent aspects of gas separation by membranes. Natu-
rally, many of the examples cited will be from our own experience, but the gen-
eral underlying principles are applicable to many membrane based gas separating
systems.
A Short History
Membranes for the separation of gas mixtures are of two very different
kinds: one a microporous membrane, the other nonporous. Microporous mem-
branes were the first to be studied and the basic law governing their selectivity
was discovered by Graham.’ When pore size of a microporous membrane is small
compared to the mean-free-path of the gas molecules, permeate will be enriched
in the gas of the lower molecular weight. Since molecular weight ratios of most
gases are not very large and since the selectivity is proportional to the square
559
root of this ratio, not only practical but theoretical enrichments achievable by
this method necessarily will be small. in order to have an efficient separation of
a gas mixture, many separation stages are required. On the other hand, since this
method of separation is based strictly on mass ratios and not chemical differ-
ences, it is the only membrane based method capable of separating isotopes of a
given compound. This is the reason it was chosen as a method to enrich uranium
in the fissionable isotope of mass 235 in the development of the atomic bomb in
World War II. This separation method inherently is very expensive-it requires a
large amount of hardware for a given amount of processed gas, the membrane
specifications are stringent (high porosity, small pore size), and the energy re-
quirements are high. The large scale utilization of such a process has been feas-
ible only because economic considerations are not of prime importance in this
application.
The other membrane-based gas separation method utilizes non-porous mem-
branes. In permeating through the membrane, the gases are separated due to
differences in their diffusivity and solubility in the membrane matrix (nor-
mally an organic polymer). Molecular size will play a role in such separations
but so will the chemical nature of the gas. Thus, conceptually very efficient sep-
arations should be possible this way. As polymer science developed, many poly-
mers were tested for gas permeabilities and indeed some with very good selec-
tivities were found. For example, many polymers are much more permeable to
polar gases (H,O, COZ, H2S, SOZ) than to nonpolar gases (O,, N,, CH4). The gases
of smallest molecular size (He, Hz) permeate more readily through polymers
than larger molecules (CH4, CZH&.’ Still it took many years to bring this basic
knowledge to practical utilization. The major problem was to make membranes
that would be both highly selective and allow high transmembrane gas fluxes,
otherwise membrane areas required would be so large as to make the technique
uneconomical. The same problem had faced desalination membranes. In the lat-
ter case, the solution appeared in the form of the Loeb-Sourirajan membrane,3
a membrane with an asymmetric cross-section, the top thin part being the selec-
tive layer and the remainder a porous, nonselective mechanical support. Still,
utilization of this technology in gas separations was not a trivial matter. Asym-
metric cellulose acetate reverse osmosis membranes have to be kept wet at all
times. Since most gas streams are essentially dry, ways had to be found to make
good dry membranes. When cellulose acetate membranes are allowed to dry,
they lose their separating power due to morphological changes caused by the
large interfacial forces involved. Various interesting approaches were suggested
to make dry membranes: for example, casting very thin films on a liquid sur-
face, and then supporting them on a porous substrate.4 Such techniques proved
to be too cumbersome to be easily scalable to industrial production. The signifi-
cant start toward practical, economical gasseparating membranes was the develop-
ment of methods of drying cellulose acetate membranes in a controlled manner.
It was found that dry asymmetric membranes are obtained with good permeabil-
ities and good selectivities in many gas separations (e.g., H21CH4, C02/CH4,
H,0/CH4) when the wet membranes are either freeze dried’ or solvent ex-
changed.6 Some inherent limitations of cellulose acetate membranes are their
sensitivity to liquid water and plasticizing gases and their limited temperature
and pressure resistance. A second breakthrough came in the form of composite
membranes introduced by Monsanto Company.’ These membranes are highly
The Separation of Gases by Membranes 561
unusual in that unlike classical composites, it is not the top layer but the one be-
low in combination with the top layer that is responsible for the membrane’s
selectivity. The membrane consists of a polysulfone hollow fiber having a small
number of surface imperfections which make it quite non-selective. By coating
with a highly permeable, non-selective polymer such as silicone, the flux through
the imperfections is diminished sufficiently to make the remainder of the mem-
brane matrix dominant in gas transport. Thus, selectivities approaching those in-
herent to polysulfone are achieved. Also, permeabiiities of the membrane are
high due to their small effective thickness which can be controlled in the mem-
brane formation process. This new technology makes it possible to fabricate thin
membranes which initially are not perfect as the effect of imperfections essen-
tially is eliminated by the coating. With the exception of some short lived pro-
jects, 8r9 Monsanto now operating as a separate subsidiary, Permea Inc., was the
first to commercialize gas separation membranes on a relatively large scale. This
apparently was an incentive to move dry cellulose acetate membranes to the
marketplace and these are being marketed today by a number of companies in-
cluding Dow, Separex, Envirogenics, W.R. Grace and others. The cellulose ace-
tate membranes are made as hollow fibers or as flat sheets which are used in
spiral wound modules. Other membranes, including immobilized liquid mem-
branes and facilitated transport, were reviewed by Matson, et al” in 1983.
Since 1983, newer systems for hydrogen separations have appeared. Ube In-
dustries, Ltd. announced the development of a polyimide resin in hollow fiber
membrane form for Hz/C1 separations. ‘*rl’ Commercial production of this mem-
brane and the engineering of skid mounted units were to begin in the fall of
1985. l3
Systems for air separations appeared in this same time period. Asahi Glass
Company, Ltd. started the production of a membrane system which is reported
to upgrade the oxygen content in air to 40%.14 The flat sheet membrane is a
fluoropolymer based composite mounted in a plate and frame unit. Air is sup-
plied at atmospheric pressure and the permeate enriched in 0s is recovered under
vacuum. Signal’s UOP Fluid Systems Division offers a similar system. Their
Spiragas@ (trademark of Signal Company) membrane for oxygen enrichment, is
an ultrathin silicone on a porous polysulfone backing in a spiral wound perme-
ator.‘s~‘6 Dow Chemical Company more recently announced the sale of the
Generon@ (trademark of The Dow Chemical Company) air separation system
to provide 95% N, for blanketing applications.” The hollow fiber membrane
used in this system is made from a polyolefin. Finally, Monsanto announced the
development of a new membrane based on crosslinked polyphenylene oxide18
which is expected to find initial application in air as well as hydrogen and car-
bon dioxide separations. More details of these systems will be given later in this
chapter.
THEORY OF GAS TRANSPORT IN MEMBRANES
Porous Membranes
The separation of gases by microporous membranes depends on the ratio of
pore size to the mean free path which is given by:
A=%! rrRTh
2P II2M
L -I (I)
where M = molecular weight
P = pressure
IJ = viscosity
R = the universal gas constant
T = temperature
Gases at 20% and 1 atmosphere have a mean-free-path of 1000 to 2000 A.

If pore size is large relative to the mean-free-path, flow (mols per unit time and
area) will be given by Poiseuille’s Law.lg
(2)
where pr & pz = the pressures on the high and low pressure sides of the
membrane
d = pore diameter
II = pore length
In such flow, energy exchange among molecules will occur within the pore, and
no separation is obtained. If, however, pores are substantially smaller than the
mean-free-path, Knudsen flow prevails:
Q = 4d (P, - P2)
(3)
31 (22rMRT)+
As this flow is molecular weight dependent, separation is expected. Ideal separa-

tion occurs when downstream pressure is negligibly small. Thus, for a gas mix-
ture:
kpI(l-x)
42 =
M21
(5)
where k is a combination of all constants in Equation 3 excluding M, and x is the

mol fraction of gas 1 in the feed, separation factor is defined as:
x/(1-x) !.!A
(JAdQ2
MI1+ and equals
[
The ideal separation factor for U 23sF&J238F6 is 1.0043. In real separations,

the ideal separation factor will not be attained because of various adverse factors
including back-diffusion, non-ideal membranes, concentration polarization at the
feed surface and surface flow phenomena. Since the actual selectivities obtain-
able are so small, it is clear that a very large number of stages will be required
for efficient separations. For a membrane cascade with undiffused gas recycled
to a previous stage, a minimum number of stages will be obtained at total reflux
(no product taken out). It can be shown*‘that this number is given by:
1
YD/ ("YD)
&-I xg/o-xg) (6)
N= [
where y & x = the mol fractions of gas 1 in the permeated and non-
permeated streams respectively
B = bottom stage
D = top stage
a = stage separation factor given as:
y/Cl-y)
= = x/(1-x)
In actual use, the number of stages obviously will be greater since reflux is
not total. Available information indicates that U.S. gas diffusion plants for U23s
oroduction utilize 10,276 stages. Total membrane area for a given separation is
given by*‘:
s = 3.914 AU& (MRT)’ . 1

(~1)~ u d (7)
P2 WP,13
where u = membrane porosity
Pr = PI/P 1
AU = separative capacity given by:
A IJ = B L (a-1>2 = 0.48L (l-p,)2 (II-l)2
where L = total molar flow rate through the membrane.
Non-Porous Membranes
As mentioned earlier, the mechanism of gas separation by non-porous mem-
branes basically is different from the one in microporous membranes. Gas mole-
cules actually dissolve and diffuse in the dense membrane matrix. Differences in
permeability, therefore, will result not only from diffusivity (mobility) differ-
ences of the various gas species but also from differences in physico-chemical
interactions of these species within the polymer, determining the amount of
gas that can be accommodated per unit volume of the matrix.
Rubbery Polymers. For polymers above their glass transition temperatures,

Tg, the permeation behavior is relatively simple phenomenologically. Solubility
of the gas in the polymer, as in liquids, follows Henry’s law,
c = s.p (9)
where C is the equilibrium sorption at partial pressure p, and S is Henry’s solu-

bility constant for the particular polymer-gas pair. Gas concentration in the
polymer thus is related linearly to the external partial pressure. Both positive
and negative deviations from this simple relationship have been observed at high
gas pressures. Gas flux through the membrane normally is found to follow Fick’s
first law.
(10)
where D is the diffusivity, defined by this equation, and C is the local concen-
tration of gas in the membrane. The diffusivity normally is concentration inde-
pendent with some deviations in the case of very soluble gases. Integration over
the membranes thickness, Q, yields:
(II)
where C1 and Cz are gas concentrations at the high and low pressure faces of the
membrane, respectively, and P is the thickness. Substituting Equation 9 into 11
yields:
Q=DS~=p~
(12)
where p, and pz are the external partial pressures of the gas on the high and low
pressure side of the membrane and P is the permeability defined by this equa-
tion. It is, therefore, given by:
P = DS (13)
Both P and D (and thus S) can be determined in a single experiment, meas-

uring the “time lag” for pressure increase on the low pressure side of the mem-
brane after high pressure had been applied the other side of the membrane. This
time lag, obtained by extrapolating the linear portion of the pressure vs time
function back to the time axis (Figure 10.1),22 is given by:
22 (14)
6
=ai
The permeability is given by the slope of the function, and solubility can be
calculated by dividing the permeability by diffusivity. For most permanent
gases, the permeability coefficient P will be pressure independent for all practical
purposes. However, for some gases with high boiling points at temperatures
lower than the critical temperature, permeability increases with pressure, reflect-
ing an increase in the solubility constant. The effects of temperature on dif-

fusivity and solubility of gas normally are of opposite signs, i.e., diffusivity in-
creasing and solubility decreasing with increasing temperature. The effect of
temperature on permeability is, therefore, a composite of these effects. Under
most normally encountered conditions, the effect on diffusivity predominates,
and permeability increases with temperature. However, below some temperature
solubility can become so high as to reverse the effect. These effects are sum-
marized in Figure 1O.2.23 The effect of temperature on gas selectivity of mem-
branes has not been studied very thoroughly. As a rule of thumb, temperature
will have a greater effect on the permeability of the gas with the higher diffusion
activation energy, i.e., the less permeable gas. Therefore, it is the usual observa-
tion that selectivity decreases with increasing temperature.
/ FVAP~R PREssuRE
OF PENETRANT
APPLIED PENETRANT PRESSURE
Figure 10.1: Typical result of transient permeation experiment.23
Glassy Polymers
Some rubbery polymers are among the most permeable polymers known.
Contrary to popular belief, however, it is not true that rubbery polymers as a
family are more permeable than glassy polymers. In fact, the most permeable
polymers reported to date have quite high glass transition temperatures.24 An
important deficiency of rubbery polymers is their low modulus. Therefore, they
will not form thin, self-supporting, pressure resistant membranes. For this rea-
son, most recent interest has been in using glassy polymers to make gas separat-
ing membranes. It was found that the solubility of gases in (and hence permea-
bility of) crystalline polymers is extremely low. Therefore, the most practically
Ttme (s)
Figure 10.2: Pressure and temperature dependence of permeability.24
interesting polymers are amorphous with high glass transition temperatures. Un-
like rubbery polymers, sorption of gases in glassy polymers does not follow
Henry’s law, and a typical sorption isotherm is convex to the pressure axis, as
shown in Figure 10.3.** This was shown to be the case with a large number of
polymers and many gases. One theory that has gained wide popularity in ex-
plaining this behavior and a broad range of other experimental data is the dual
sorption theory.22~25~26It postulates the existence of two types of sorption sites
in the glassy polymer. One accommodates mobile gas molecules following
Henry’s law. The other consists of Langmuir type sites with much less mobile
sorbed species. The Langmuir sites have been associated with fixed voids or ex-
cess free volume in the glassy state and correlated with the abrupt change in
slope of the specific volume vs temperature curve as one passes the glass transi-
tion temperature, Figure 10.4. *’ The two sorbed gas populations are in rapid
equilibrium throughout the membranes. This behavior is described by the
following equation:
c = CD + CH = sp +
5i- * b (15)
1 + bp
where CD = solubility in the Henry type sites

C,, = solubility in the Langmuir type sites
c;, = hole saturation constant
b = hole affinity constant
25 ‘C
1D
1 I 1 I I I
0
0 2 4 6 0 10 12 14 16 16 20 22 24 26 26 30 32 34 36
Pressure (atm. abs.)
Figure 10.3: Solubility of methane in polystyrene.23
Figure 10.4: Specific volume dependence on temperature in the region of the

glass transition.26
Sorption behavior at low and high pressure is consistent with the model: At
low pressures (bp < I), Equation 15 reduces to:
C = (S + C;I b) p
a modified Henry-like behavior. At high pressure (bps I), Equation 15 becomes:
c = sp + c;I (17)
indicating saturation of the Langmuir sites. Thus, C is a linear function of p at

both low and high pressure with a nonlinear region in between. This corresponds
well with the results as shown in Figure 10.3. These mathematical relations allow
the determination of the various dual sorption parameters. Thus,, from the pres-
sure dependence of C at high pressures (Equation 171, S and CH can be deter-
mined from the slope and intercept, respectively. Writing the Langmuir part of
the sorption equation:
C;I bp
cH= -
1 + bp (18)
rearranging to
P/CH = l/C; b + p/C;,
(19)
and plotting p/C, vs p, one can determine CL and b (CL is obtained as the dif-
ference between total sorption and Sp).
Total solubility of gases in glassy polymers is higher than in rubbers, which
is consistent with the apparent existence of an additional sorption site in such
polymers. The temperature dependence of Henry’s law constant
S = See -“DiRT
and the dependence of the Langmuir constant:
b = hoe MH’RT (21)
enable the determination of the sorption enthalpies into the various sites. It was
found that the Henry-type sorption enthalpy, AHD, is of the same order as that
in rubbers, whereas the enthalpy for sorption in Langmuir sites, AHH is appre-
ciably higher. This accounts for the empirical finding that gas sorption in glassy
polymers is substantially more exothermic than in rubbers, and it is explained
in dual sorption theory by the sorption of gas into preexisting sites rather than
sites that need to be formed by polymer molecular reorientations (as in rubbers).
While the permeability of most rubbery polymers to most gases is pressure
independent, the permeability of glassy polymers usually is observed to decline
with increasing pressure. This is accounted for by the dual sorption theory if
one assumes that both the Henry and Langmuir type sites contribute to per-
meability. ” Thus , a modified one dimensional Fick’s law is:
Q = -DD$j -Do 2 (22)
where DD and DH are the diffusivities of gas molecules associated with the
Henry and Langmuir environments respectively. Solution of this equation yields
P= SDD [l +Ebp] (23)
where F = DH/DD and K = CL b/S.

The time lag also takes on a more complicated form:
0 = $ [I + f (K,F,bp)l
where the function f is too complex to reproduce here.

For very high values of bp, such as at high pressures, when Langmuir sites
are saturated and thus do not participate in transport, Equation 23 reduces to
the simple Henry form.13
At low pressure, the following limiting forms are obtained:
P = SDD + b C;r DH
6 = Q* (l+K)
6D (l+FK)
These relationships were demonstrated for many gases in various polymers.

In spite of its wide acceptance and success in representing a large amount of ex-
perimental data, the dual sorption theory has come under attack recently. One
reason is that it has never been demonstrated conclusively that two sorbed gas
molecule populations actually exist in polymers. An alternate model was sug
gested-the matrix model. 27 In this model, it is assumed that only a single popu-
lation of sorbed gas molecules exists and that the gas changes the properties of
the polymer matrix over the entire pressure-concentration range in a continu-
ous manner. Mathematical expressions were developed to describe the depend-
ence of solubility, apparent diffusivity, permeability and time lag on gas concen-
tration. These are at least as successful as dual sorption expressions in describing
experimental data. Thus, the concentration dependent solubility constant is
given by:
S=i-----
%I
- (Yc (27)
where C is the equilibrium concentration of gas in the matrix and (Y is a constant.

The concentration dependent diffusivity is given by:
D = Do (1 + PC)
/?, the “interaction parameter” is positive for a plasticizing gas and negative if the
gas has antiplasticizing effect. By solving Fick’s first law with the appropriate
expression for the diffusivity, one obtains the concentration dependent permea-
bility as:
P=DS (291
00 +t-kg-
The expression for the concentration dependent time lag is:
22 10 + 15jX + 16BW (30)

B=a
10 (1 + jIC)3
The fit of these expressions to experimental results is very good. At low

pressure regimes, the fit was shown to be even better than that of dual sorption
expressions. Except for these regimes, the two models seem to do equally well in
describing sorption and permeability data. Concentration dependent diffusivity
and permeability have been considered before mainly for vapors. The new aspect
of the matrix model is that it broadens these effects to fixed gases. The impor-
tant difference between the matrix and dual sorption models is in the physical
picture they convey of gas transport and interaction with the polymer. Addi-
tional experimental evidence will be needed to determine the preference of these
different physical representations.
The theories described above, though successful at correlating many experi-
mental data, are phenomenological in nature. Attempts have been made also to
explain gas diffusion in polymers on a molecular basis. One of the more recent
theories is that by Pace and Datyner” which elaborates on earlier theories of
Brandt29 and DiBenedetto.sa Gas molec u les are viewed as being trapped in chan-
nels formed from polymer chains over short distances. Longitudinal diffusion
along these channels is very rapid with low activation energy. The true activa-
tion energy for diffusion is that associated with transverse diffusion out of the
molecular cage which depends on polymer chain reorientation. The theory pro-
vides an expression for the activation energy for diffusion based on first princi-
ples. The expression predicts an apparent activation energy (that obtained from
an Arrhenius plot) which is independent of temperature above and below Tg,
with the value above Tg higher than below. This general behavior is indeed ob-
served experimentally. The theory also derived an expression for the diffusion
constant; however, it contains a parameter, the mean-square-jump displacement
of the diffusant, which cannot be determined as yet.
ENGINEERING ASPECTS
The efficiency of a gas separation process can be described by essentially

two parameters: the purity of the product gas and the fraction of that gas in
the feed recovered as product (the recovery). These parameters in turn will be
determined by the membrane’s intrinsic properties-its permeability and selec-
tivity and by operating factors such as total and partial pressures on the feed and
permeate sides, feed flow rate and pressure drop on either side of the membrane.
These factors will determine the membrane area and compression energy neces-
sary for a given separation, and only with all these parameters can an economic
evaluation of the process be made.
The basic engineering analysis of gas separations by membranes was made
by Weller and Steiner31r3’ long before good, practical membranes were devel-
oped, but it is still useful as a good approximation. Assuming a binary gas feed
and laminar flow on both high and low pressure sides of the membrane, permeate
composition at any point will be determined by the transport rate of the two
gases at that point:
1
-dnA - P2 dnA
(31)
dnA + dnB
1
-dng = PB dA p1 “B - P2 dnB
(32)
dnA + dnB
C “A + “B
where PA & PB = the membrane’s permeability to gases A and B

respectively,
PI3 P2 = total pressures on the high and low pressure side re-
spectively,
“A & n0 = the numbers of mols of gases A and B flowing per

unit time on the high pressure side, and
dnA & dnB = permeation rates.
An analytical solution to these equations giving the molar ratio of compo-
nent B leaving and entering the membrane stage is:‘e
f
“B to -01 + c to - c
Qn 7; = RQn to - B/A + UQn + TQn - (33)
“B tf - B/A tf -cr+c tf - c
where 01 = pA/p0
A = [(I -cu)p,/pl+al/2
B = - AC + al2
C = - [(l - dp*/P1 - II/2
R = 1/(2A - 1)
U = [a(A-l)+Cl/(2A- I) (o/2-C)
T =i l/(1 -A - B/C)
t = -Ai + [A2i2 + 2Bi + C21X
i =
nAh
Once nfg (rate of gas B leaving the high pressure side) has been determined,
bore composition can be calculated by mass balance:
(34)
where y is the mol fraction of gas A in the permeate, o and f denote entering and
leaving, respectively. The recovery of gas A will be given by:
Obviously, the higher the recovery the lower the purity of the permeated
gas. In the special case of zero recovery or complete mixing (and therefore, con-
stant shell and bore compositions along the membranes), an especially simple ex-
pression is obtained by dividing Equation 31 and 32:
(35)
where y = the mol fraction of gas A in the permeate

X = the mol fraction of gas A on the high pressure side
01 = the separation factor PA/Pg
If, in addition, bore pressure is zero, ps = 0, the selectivity and the mol
fractions show the simple relation:
y/ (1-x) = c1
(36)
x/ (1-y)
Equation 35 clearly shows that product purity is directly related to membrane

selectivity. Solving the quadratic Equation 35 for y yields:
y = -B+[B2+4(1-CY) a(pl/pz)x]’ (37)
2(1-a)
where B = (o - I) (PJP*) x + PI/P2 + a- 1
Thus, an increase in total pressure ratio pl/ps will improve purity for a given 01.
The various effects of pressure ratio and membrane selectivity are shown in
Figure 10.5.33 It will be observed that under certain conditions the pressure ratio
is of even greater importance than intrinsic selectivity. Thus, with a feed gas rela-
tively poor in the more permeable component, and an (Yof about 40, further in-
crease in selectivity has little effect on product purity, whereas an increase in
P, = Absolute feed gas pressure
60~~11r~1111~~‘r~“lrl’lll~‘ll’l’r’l”ull’li’1
0 10 20 30 40 50 60 70 80 !
Separation factor
Figure 10.5: Product purity as a function of feed composition and membrane

selectivity.33
pressure ratio has a substantial effect over the whole selectivity range. The effect
of pressure ratio is much smaller if the feed gas is already rich in the more per-
meable component. It is seen that with any feed composition the effect of mem-
brane selectivity on product purity reaches a plateau above some a! value. There-
fore, research effort to develop super-selective membranes always should be con-
sidered with this trend in mind.
One of the main determinants of process economics is the membrane area
required for a given separation. The exact solution, obtainable by numerical
methods, is:
J
.O
1 (38)
antiln [n,(i)/ni] di
S'-- p;
PB [f(i)-11 [pl(l/(i+l)) - p2(ll(f(i)+l))l
if
where i = nAfn0
f(i) = Ai - C + [A*i* + 2Bi + C*]x (see Equation 33)
A good approximation for the area can be obtained by averaging the solu-
tions of Equation 35 for inlet and outlet compositions. Thus, a permeate mol
fraction at the inlet, yo, can be determined from:
_?!I_=. PI x0- P*Y”

(39)
l-y” PI (l-x0)-p, (1 -y”)
(x0 is the mol fraction of gas A at the feed inlet). If it is desired to obtain AnA
mols of gas A per unit time, a corresponding area will be:
so =
p&lx0 - p2y”)
A similar expression can be written for the stage outlet, and the average of
the two areas is taken as the approximate required area. The area is thus seen to
be inversely proportional to fast gas permeability and partial pressuredifferential.
It should be realized that the membrane contributes only a part (sometimes
a small part) of total capital cost of the separation system. Compressors, pre-
treatment facilities and mounting may amount to more than membrane cost.
Pretreatment to remove contaminants harmful to the specific membrane can be
especially costly. It is, therefore, desirable to have membranes that are highly
environmentally resistant or that can be operated under conditions (e.g., high
temperature) in which the solubility of the contaminants in the membrane is
low. Compression may be a major factor in system operating cost. Therefore,
membranes with high permeabilities are desired to minimize the need for high
driving force and to minimize membrane area. If membrane selectivity is too low,
multistaging will be necessary. As a rule, this will increase both capital and op-
erating cost due to the additional membrane area required and the recompres-
sion between the stages.
Gas separation by membranes will always have to compete with other sepa-
ration processes such as cryogenics, absorption and adsorption systems. Mem-
branes usually are less competitive in very large scale operations where the fast
gas is less than about 20% of the feed gas, unless the slow gas is the desired prod-
uct. Membranes also are not usually the method of choice when extremely pure
product gas is required. Membranes do, however, have distinct advantages in
small to medium scale operations, in situations where gas is available at pres-
sure, in situations where high recovery is paramount, and in applications where
simplicity and minimal maintenance are of prime importance (such as in remote
locations). Membranes are very well suited for applications in which the non-
permeate is the product of interest, since it is obtained at pressure. Examples
are acid gas removal from natural gas and gas dehydration.
PREPARATION OF MEMBRANES FOR GAS SEPARATION
Permea’s (Monsanto) Composite Membrane
Membranes to be used in practical gas separations have to satisfy many re-

quirements; they have to exhibit high fluxes so as not to demand excessive mem-
brane areas for large volume gas processing. They must have good selectivities
in order to offer product gas streams of high purity. They must be able to with-
stand the high gas pressures often encountered in natural and industrial streams.
They must withstand reasonable concentrations of contaminants often accom-
panying the gases of interest (e.g., H,O, H2.S. hydrocarbons).
It was the combination of these stringent requirements which impeded the
development of useful gas separating membranes for so many years. Obviously,
one of the most difficult requirements is the achievement of high permeability

and high selectivity simulataneously. A similar problem had hampered the de-
velopment of reverse osmosis membranes. It was solved brilliantly in the asym-
metric Loeb-Sourirajan membrane, having a thin selective layer supported by a
much thicker porous. non-selective matrix. In this membrane. the fact was ex-
ploited that whereas permeability is thickness dependent. selectivity-being a
ratio of permeabilities-is not. The problem with gas separation membranes is
even more severe. Reverse osmosis membranes can show good salt rejection even
if slightly porous, by virtue of electrostatic interactions (non-availability of free
water to solvate ions inside small pores). Most common gases are of very small
effective diameters (2 to 5 A) and mechanisms such as the above to increase se-
lectivity are not available. Also. permeabilities of gases in polymer matrices are
usually very low,
10-12- 10-8 Scc -cm
cmz-sec-cmlIg
Therefore, membranes for gas separation are extremely sensitive to the existence
of even a relatively few small imperfections, due to the very high permeability
of such holes relative to the imperfection-free areas. The solution developed by
Monsanto is unique in that it circumvented the need to make a perfect memo
brane by devising a special coating procedure.34 On coating the membrane with
a thin layer of a highly permeable, non-selective polymer, the permeability
through the holes is reduced sufficiently to make the permeation through the
matrix predominant (Figure 10.6). Performance close to that expected from in-
trinsic (dense film) properties is thus achieved. The materials presently used in
Permea's (a Monsanto Company) commercial membranes are a polysulfone as
the matrix and a silicone rubber as a coating. However, the method is general
and any matrix-coating combination with the appropriate permeability ratio can
be used. A numerical example will illustrate the utility of such membranes in
the separations of CO2 from CH4.
PM COMPOSITE -SCHEMATIC PM COMPOSITE -ELECTRICAL CIRCUIT ANALOG
Figure 10.6: Schematic representation of Monsanto's composite membrane.

Cross-sectional view on left, electrical circuit analog on right.34
Assume a polysulfone fiber with an effective thickness of 1000 a and a

surface porosity of 10% (ratio of pore area to total area). The fiber is coated
with a silicone at a thickness of 1 pm. The intrinsic permeabilities of polysulfone
and silicone are:
See - cm
PC02 (polysulfone) = 6 x 10-l'
cm2-set-cmHg
See - cm
PCH.j (polysulfone) = 0.2 x lo-l0 cm2_sec_cdg
See - cm
PC02 (silicone) = 3500 x 10-l'
2-sec-cmHg
GFC - cm
PCHl (silicone) = 800 x 10BfO
cm2-see-cmHg
Remembering that the permeability term P/Q (where Q is the effective mem-
brane thickness) is reciprocal resistance, the total COz resistance of the mem-
brane will be:
II
-1 -1
(PCOa) Si x porosity + (PCO*)PS
LPS
thus
(P/g) C02=~50;0;10-10) -' +(3500 * lO-'O . ;;Tf + 6 . 10-l‘,J-j-I

= 61 . lo-6 zzc
2-sec-cmQ
Similarly for methane
foe ;,~rlo)-l +(,oo * 10-10 l ;g + 0.2 lo-l~-l]-~

(P/Q) CH4
= 1.99 * lo-6 f;;_sec_cmHg
the separation factor,
(P/QKO, = 3o.6
o = (P/Q)CH,,
is essentially identical to the intrinsic (dense film) separation factor of -30.

The uncoated membrane has a much lower selectivity: assuming cylindrical
pores, the molar Knudsen flow through the pores will be given by:
Q =4d lc?_czz
31 (2lrMRT)i
where d = Pore diameter

II = Pore length
M = Molecular weight
R = Gas constant
T = Absolute temperature
If the pressure differential is 1 atmosphere, CO* flow through the pores in our
example will be:
112 x 10-e =csSTP

. cm membrane area-set-atm
methane flux will be
186 x 10-e ,‘;s _ set

- atm
The corresponding fluxes through the polysulfone matrix will be
6 x lo-lo
for CO2: =
1000 x 10 -8 x 76 4560 x lo-" z& _ sec - atm
and for methane
1000 * .10-10
0.2 10-s 76 = 152 x 1O-6 &,c_atm
Adding reciprocal permeabilities as resistances, overall selectivity will be
4672
CIC02/CH4 = - = 13.8
338
The coating thus eliminates the detrimental effect of the imperfections without
affecting the flux in any major way.
It can be shown that the coating technique will give an almost constant per-
formance over a wide range of surface porosity (Figure 10.7). Only at relatively
high porosity (10T3 or higher) will flux of the slower gas through the plugged
pores become large enough to adversely affect the selectivity. Also, the tech-
nique is not very sensitive to coating thickness and only at thicknesses of 1 pm
or greater will the coating contribute significantly to the total resistance (thus
lowering selectivity). It is this flexibility and the freedom from very stringent
and expensive process control which makes this technology so attractive.
A number of advantages of this kind of composite membrane are note-
worthy. It is not sensitive to water and, therefore, feed streams do not have to
be dry. Its temperature resistance is good and, therefore, operation at elevated
temperatures is a simple and inexpensive way to minimize the effect of con-
taminants and to increase permeability. The pressure rating is excellent, and the
RM COMPOSITE
-- -___---- POROUS SUESTRATE-- - -- --____

L 1
10-10 10-8 10-6 10-4 10-2
SURFACEPOROSITY (As/A21
Figure 10.7: Effect of surface porosity on the separation factor (H&O) for a
porous substrate and Monsanto’s composite membrane.Y1
membranes perform well for extended periods in high pressure feeds such as
those encountered in ammonia plants.
The composite membranes were made into successful systems thanks to var-
ious additional developments, including appropriate chemical and pressure resis-
tant module seals35l36and shell design. These membranes were first used in house
and by now have found numerous applications throughout the world.
Dry Cellulose Acetate Membranes

As stated previously, asymmetric cellulose acetate reverse osmosis mem-
branes are kept wet at all times. Uncontrolled drying brings about a loss of se-
lectivity. This results from the large interfacial forces developing during drying
of the small pores of this hydrophilic polymer. Such uncontrolled drying can
occur on exposing a wet membrane to a gas stream. Therefore, techniques had
to be found to dry cellulose acetate in a controlled manner before the mem-
branes could be used to separate gases. In 1968, it was suggested that drying by
solvent exchange could lead to a good gas separating membrane.6j37 The pro-
cedure called for soaking the wet membrane in isopropanol for half an hour
followed by toluene for several hours. Water is thus removed by dissolution in
the alcohol and all polar solvent is removed before exposure to air. It was soon
learned that surfactant treatment or freeze drying would achieve a similar re-
sult.3*,39
Thus, a freeze dried membrane showed a He permeability of
3 - 10 x 10-5 zz2 _ set

- cmlrg
and (Y He/N, selectivity of 34-46 vs 97 for the dense film. Apparently, a small
number of defects were present in the membrane. Improvements in the drying
procedure were then developed such as carrying out the solvent exchange at low
temperature.40 For example, a membrane treated at O’C with n-butanol fol-
lowed by n-heptane showed (P/Q) He of
and a HelCH, of 62. Heat treatment at 95’C following the drying was found to
be beneficial, increasing He/CH4 selectivity to 75.
Cellulose acetate membranes have been suggested for hydrogen recovery,
C02/CH4 separations, gas dehydration and for nitrogen enrichment.41A3
Composite Polyimide Membranes

Ube Industry, Ltd. introduced a polyimide membrane which is reported to
be made from a condensation polymer of biphenyltetracarboxylic anhydride
and an aromatic diamine such as 4,4’-diaminodiphenyl ether or a mixture of
two diamines.44A6 The comparison in Table 10.1 indicates the intrinsic permea-
bilities for Hz in these polyimides are similar to polysulfone, and the Hz/CO se-
lectivities are two times greater. Likewise the 02/N2 selectivities are twice those
of polysulfone, but the O2 permeabilities are only one-half the polysulfone val-
ues. These polyimide materials are good candidates for gas separation mem-
branes owing to their high selectivities, high insolubilities, and high strengths
and glass transition temperatures.45r46
Asymmetric membranes are made from solution in the form of a hollow fi-
ber, but the process used to form a thin, pore free dense layer on these hollow
fibers is not disclosed.45j46 However, US patent 4,440,64312 describes a unique
process for producing pore-free polyimide composite membranes. An asym-
metric polyimide porous substrate is prepared from solution. When fully imi-
dized, the substrate is insoluble. The substrate can now be coated with a poly-
amic acid from dilute solution (-1%). When fully imidized, the resultant poly-
imide coating becomes the separating layer. This process allows use of the same
or different polyimides for the substrate and the separating membrane. While
the examples in the reference describe preparation of flat sheet membranes, this
process could be used to prepare hollow fiber membranes.
A recent report 49 indicates the permeabilities of the composite hollow fiber
membranes made by Ube may be considerably lower than those of polysulfone
when compared at the same temperature. The Japanese literature suggests higher
permeabilities than given by Reference 49. It is not known which is the case for
the membranes actually used in the separators. The polyimides are reported to
be capable of sustained, higher temperature operation to give permeabilities
comparable to those of polysulfone while still retaining adequate selectivity.45t46
The cited references indicate the hollow fibers are assembled and installed in
a module of a design very similar to that employed by Permea. The success of
sustained, high temperature operation, which may be needed for permeability,
will depend on the development of heat resistant end seal materials. The materia-
als must not only withstand high temperatures, but also survive the effects of
Table 10.1: Intrinsic Permeabilities and Ideal Separation Factors for Selected
Polymers
Polymer (Reference) H2/CO Hz/CH4 m C02/CH4(b) g&j2
;zg;;;:;n;ti) G& 47) 13

9 6 0.5
1 40
76 -200
60 -200
72 25-30 11
6
Poly(4-methyl’pentene-1) (48) 136 93 32 - 17 4

Crosslinked PPO (18) 22 9 2.3 37 55 56 22 6
Silicone Rubber (51) 550 271 501 - 0.1 2 0.3 2
Cellulose Acetate (64) 12 6 1 40 60 70 30 5.5
(a) Units are cm3 (STP)-cm/cm* set cmHg at temperatures of 25“-35Y.

(b) Low pressure (<20 atm). The separation factor of thispair varies with pressure.
(c) Product of Union Carbide Corp.
temperature cycling and stream contaminants to prevent failure by cracking.
Polyolefin Membranes for Air Separation

The membrane is made from a polyolefin. Based on work Dow did for the
Federal Aviation Agency and reported performance capabilities, it is believed
that poly(4-methylpentene-1) is used.” A comparison of the intrinsic permea-
bilities of this polymer with polysulfone and polyimide is given in Table 10.1.
Poly(4-methylpentene-1) is highly permeable, but it has low selectivities as in-
dicated by the Oz/Nz pair data and the 01 of 17 for H2/N2. While the melting
point of this crystalline polymer is high, 235”C, the glass transition tempera-
ture” is much lower than polysulfone or polyimide necessitating operation of
a membrane system at temperatures close to ambient to maintain the best se-
lectivities.
The membrane is made in hollow fiber form. It is melt spun to nominal di-
mensions of 40 pm outside diameter and 30 pm inside diameter to give a wall
thickness of about 5 E.cm.The wall thickness of a melt spun membrane is much
thicker than the effective separating layer of an asymmetric membrane which is
on the order of 0.1 pm. However, the permeability of the polyolefin is high and
the small dimensions allow use of a very large membrane area in a compact unit
which together compensate for the large membrane thickness. A disadvantage of
small dimension fiber is the large pressure drop in the bore side due to capillary
flow which limits the length of the fiber in the separating unit and therefore the
area. This effect is accommodated by appropriate module design. This mem-
brane is well suited for air separations, but the low selectivities will prevent its
application to hydrogen in carbon dioxide separations.
Other Membranes
Signal’s UOP Fluid Systems Division Spiragas (a trademark of Signal Co.)
membrane is prepared from an ultrathin silicone material on a porous polysul-
fone flat sheet. A spiral wound module is prepared from this membrane. Sili-
cones, typically, have very high permeabilities and low selectivities.2t51 For this
reason, Spiragas is sold for oxygen enrichment applications rather than a system
to produce nitrogen of greater than 95% purity.
The Asahi Glass membrane also is reported to be a composite based on a
fluoropolymer. Flat sheets are assembled into a plate and frame module design
rather than spiral wound.
Monsanto announced the development of a crosslinked polyphenylene ox-
ide membrane early in 1985.18 While the intrinsic permeabilities of this material
are lower than polyphenylene oxide, they are higher than those of polysulfone,
Table 10.1. The membrane is made by brominating PPO followed by base cross-
linking. The transport properties of this material can be manipulated by altering
the bromine content and degree of substitution in the ring or methyl groups.
The bromine on the methyl group reacts readily with bases affording the oppor-
tunity for crosslinking and further transport property modification. This ma-
terial is spun into hollow fiber membranes using Monsanto technology. Typi-
cally, the fiber is treated with aqueous ammonia followed by coating with a
silicone rubber.7f’* The membrane is reported to be more resistant to stream
contaminants and due to the higher permeabilities is more productive than poly-
sulfone.
Membrane Testing and Evaluation

Since gas separation by membranes is a relatively recent development, it is
still quite common to base selectivity determinations on pure gas tests. This will
sometimes give a reasonable estimate of membrane performance but can often
lead to erroneous results. The reason is that in many cases one gas in a mixture
will influence the transport of the other gases. Usually selectivity measured with
mixtures will be lower than that obtained from pure gas measurements.
Also from the discussion in Chapter 1, it is clear that membrane perform-
ance will be strongly influenced by testing conditions, i.e., pressure and tempera-
ture. Therefore, it is most advisable in evaluating membranes to test them under
conditions as close as possible to actual operating conditions.
Future
Present day membranes are capable of separating mainly inherently “fast”
gases (e.g., Hz, COZ) from inherently “slow” gases (e.g., Nz, CH4). Current tech-
nology is not sufficiently refined to efficiently separate different traditionally
slow gases (such as N, from CH4, hydrocarbons from each other). Such separa-
tions can, in principle, be achieved by facilitated transport. In this process, imi-
tating the function of biological membranes, the membrane contains a com-
pound interacting specifically with only one of the feed constituents. With ap
propriate binding constants and binding kinetics, very efficient separations can
be obtained. To date most work in this area has been confined to liquid mem-
branes (thin liquid films, liquid in liquid microcapsules or liquid impregnated
solid films). Transport facilitation was demonstrated for 01, Nz, COZ, HzS and
ethylene (extensive work was done on ion transport in liquids). The most prac-
tically advanced facilitated gas transport system has been the ethylene/ethane
separation studied at Amoco. ‘* This separation, based on aqueous silver ion, was
scaled up to pilot plant size. Technical problems that have not yet been solved
in these systems are associated mainly with the need for water in the membrane
and hence in the feed streams. It is difficult to prevent the membrane from dry-
ing at all times, especially at high pressure differentials. Once such problems are
solved in an economical way, new, previously impractical gas separations should
become possible.
APPLICATIONS OF GAS SEPARATING MEMBRANES
Within a relatively short period after their commercialization, membranes

for gas separation have been utilized in a wide variety of applications. Among
these applications are recovery of hydrogen from purge streams such as encoun-
tered in ammonia plants, retrofit chemical plants and from hydroprocessors in
refineries. Membrane systems have been introduced into oil fields for CO2 re-
covery from well-head gas in enhanced oil recovery, and they have been used to
separate oxygen from nitrogen in air. Also, they have been utilized in combina-
tion with other recovery systems, such as cryogenic and adsorption, to both re-
duce cost and increase efficiency.
These applications will be discussed in detail in the remainder of this chap-
ter. The cross section of a PRISM@ (trademark of Monsanto Co.) separator is
shown in Figure 10.8. It is a simple, sturdy structure providing feed inlet and
outlet for permeate and non-permeate. Feed and permeate flow is countercur-
rent and the fibers have one end plugged near the non-permeate outlet.
Hydrogen Recovery
Wasted hydrogen from the inert purge of reactors can be recovered by
membrane separator systems. This hydrogen had normally been either flared
or burned for fuel, and its chemical value was lost. The ammonia purge gas re-
covery system was developed at the Monsanto ammonia plants in Luling, Louis-
iana. Figure 10.9 shows a two-stage pressure operation which maximizes re-
covery while minimizing recompression costs. Systems can now be designed
with hydrogen recoveries as high as 95% at a purity of 88%. Operation at dif-
ferential pressures up to 1,650 psi has been demonstrated.
Figure 10.8: PRISM@ separator.

Between 1979 and 1985, forty-five PRISM@ separator recovery systems

have been successfully implemented on ammonia purges around the world. As
a result of hydrogen recovery, an incremental ammonia production of 3 to 5%
can be achieved. Alternatively, at constant ammonia production energy savings
amount to 0.6 to 0.9m BTU/ton ammonia. For a typical 1,150 ton/day plant,
these improvements translate into $3i/year energy savings.
Figure 10.9: Hydrogen recovery from ammonia plant purge.53
Prior to entering the hollow fiber PRISM@ separators, the feed gas is passed
through a water scrubber. This serves a dual purpose of recovering ammonia and
extending the fiber life. The ammonia scrubber pretreatment alone contributed
about $200,00O/year to the equipment payback in recovered ammonia.” Typi-
cally, the recovered hydrogen is recycled back to the ammonia synthesis com-
pressor. However, if a source of high purity hydrogen is needed within the com-
plex, some or all of the gas from the first bank of separators can be used for that
application. If higher purity hydrogen is desired, the permeate from the first
bank can be processed by an additional PRISM@ separator to produce, for ex-
ample, 99% hydrogen.
In methanol synthesis purge recovery, a water scrubber is also used with a
similar purpose, and it too pays for itself in recovered methanol. The meth-
anol/water mixture is simply sent to the existing crude methanol distillation
column. Hydrogen recovered from this purge can result in energy savings or if
additional carbon oxide is available, it can be used to obtain increased methanol
production. PRISM@ separators have operated on stoichiometric as well as non-
stoichiometric H,/(CO)x ratio methanol plants at 75O and differential pressures

up to 1,000 psi.
As a major user of expensive and often scarce hydrogen, hydrocrackers are
also prime candidates for hydrogen use optimization. This enables not only the
conservation of hydrogen by purge recovery, but also high-pressure loop modi-
fications to adjust hydrogen partial pressure in the reactor. A simplified process
flow diagram of a typical hydrocracker is shown in Figure 10.10. The oil and
hydrogen feed (including recycle hydrogen) are charged to the reactors at the
desired space velocity. The effluent liquid and vapor pass through a heat ex-
changer to a high-pressure separator, where the recycle gas is separated from
the liquid. The recycle gas is returned to the reactor inlet by a recycle com-
pressor. The liquid is let down in pressure, and hydrogen and light hydrocar-
bon gas are released in a low pressure separator. The liquid product from the
low pressure flash is sent to fractionation. In the reactor, hydrogen partial
pressure is reduced by consumption through chemical reaction, by dissolved
hydrogen being carried out in the product, and by the formation of light hy-
drocarbons which dilute the remaining hydrogen. There are several ways of
maintaining reactor hydrogen partial pressure within the desired range. The
most common involve the addition to the high pressure loop of either a purge
or an oil absorber. Hydrocarbons, which dilute the reactor hydrogen, can be re-
moved by purging a part of the gas stream coming from the high-pressure sepa-
rator. A purge, however, is not selective; for every mol of hydrocarbon purged,
typically four mols of hydrogen are lost. Because of the expense of replacing
this lost hydrogen, high-pressure purge rates are usually kept to a minimum.
Figure 10.10: Hydrocracker.55
The most common alternative to the high pressure purge is oil scrubbing.
This involves an absorber vessel in the high-pressure loop, a low-pressure flash
drum, and a high pressure pump to return the lean oil to the absorber. The
principal advantage this system has over purging is hydrogen conservation. The
oil scrubber flash-drum offgas typically contains one mol of hydrogen per mol
of hydrocarbon; thus absorbers lose only about 25% as much hydrocarbon as
high-pressure purges. The main disadvantage of the absorber systems is their
cost. High operating pressures makes the mechanical components expensive, and
pump power requirements often exceed 1000 kW.
The addition of a PRISM@ separator to the recycle loop will effectively re-
move gases such as methane from the process. A simple flow diagram of a hy-
drocracker with the addition of the PRISM@ separators is shown in Figure
10.11. A recycle gas slip stream is fed to a liquid separator, where any entrained
liquids and mist are removed. The gas is then heated to 150’ to 180°F (339’to
355°K) prior to entering the PRISM@ separators. Hydrogen which permeates
the fibers is recovered and returned to the make-up compressor. With about 90%
of the hydrogen removed, the non-permeate is primarily hydrocarbons and can
be used as fuel gas. Because of the low hydrogen content of the non-permeate,
three to four mols of hydrocarbon are rejected for each mol of hydrogen lost.
BY comparison, simple purges lose 12 to 15 times as much hydrogen.”
Other refinery applications in which PRISM@ separators have been success-
fully used are naphtha hydrotreaters, middle distillate hydrotreaters, cat crack-
ers, and toluene hydrodeal kylation.56r57
The Ube system was designed to be capable of handling the applications de-
scribed above for PRISM@ separators. Compared at the same temperature, it is
reported that the permeabilities of the Ube composite polyimide hollow fiber
membranes may be substantially lower than those of polysulfone,49 but with
two times the selectivities. The lower permeabilities are overcome in two ways.
The Ube module contains 9000 (m2/m3) membrane area per unit volume com-
pared with 4000 (m2/m3) in a PRISM@ separator. Since the separator units are
approximately the same size (from Permea, Inc. and Ube product brochures),
the Ube membrane must be smaller in diameter than is the PRISM@ separator
membrane in order to increase the unit membrane area. In transporting large
volumes of Hz, small diameter hollow fiber membranes often lead to a large
bore pressure drop on the permeate side. The pressure drop is due to flow of
gas in a long capillary and is inversely proportional to the inside radius of the
hollow fiber to the fourth power, i.e., Poiseuille’s law. Pressure buildup on the
permeate side of the membrane results in a reduction of the differential partial
pressure driving force with attendant reductions in product recovery and purity.
Thus, fiber dimensions are important in designing a system for a given separa-
tion, In practice, the Ube module most likely will not function as if the unit
membrane area is more than twice that of a PRISM@ separator due to differ-
ences in fiber dimensions. The penalty, however, will depend greatly on oper-
ating conditions.
The polyimide membrane is reported to be capable of operating at tempera-
tures up to 15O’C compared to upper limits of IOO’C for PRISM@ separators and
60°C for cellulose acetate.& Permeab’III‘t’res increase with temperature while selec-
tivities normally drop. The second way to overcome the lower permeabilities is
to operate the Ube system at higher temperatures than possible with polysulfone
or cellulose acetate. Even at the high temperatures, the polyimide selectivity will
remain high enough for the abovementioned hydrogen separations. Thus, one ex-
pects the Ube system to be competitive with PRISM@ separators for many hy-
drogen applications.
The cellulose ester systems have some application in hydrogen separations,

but are more limited in capability due to temperature limitations. Operation at
higher temperature minimizes the need for stream pretreatment to remove com-
ponents which may attack the membranes. The PRISM@ separators and the Ube
systems are superior to the cellulose esters in this respect.
Carbon Dioxide Separation

Both hollow fiber and spiral-wound membrane systems have been utilized
to recover carbon dioxide in sour gas purification and in enhanced oil recovery.
Removal of hydrogen sulfide (HZS) and carbon dioxide (CO*) from natural
gas is an ideal application for membranes in that both H,S and CO* permeate
through membranes at a much higher rate than methane, enabling a high re-
covery of the acid gaseswithout significant loss of pressure in the methane pipe-
line product gases.4143f 584o
Membrane gas separation by PRISM@ separators offers a low cost processing
alternative to conventional COs removal processes61t62as shown in Table 10.2.
Once the needed partial pressure driving force is achieved by compressing
the feed gas, no further expenditure of energy is necessary to separate the
co*.63
Separex’s,4143 Grace Membrane Systems cellulose ester,60 Envirogenics
GASEP@ (trademark of Envirogenics Company)64 cellulose triacetate spiral-
wound membranes, and Dow cellulose acetate hollow fibers are used to pro-
duce a salable product from sour gas streams.
Table 10.2: Product Purity and Recovery
HOT POTASSIUM CRYOGENIC PRISM@

SEPARATORS SYSTEM CARBONATE SEPARATION SEPARATORS
CO2 Product Stream
COP Recovery % 96.9 93.0 96.9

CO2 Purity % 99.8 95.1 95.5
Overall Hydrocarbon Stream
Hydrocarbon Recovery % 99.5 81.0' 83.5
Hydrocarbon Gas Product
CO2 % 2.0 2.0 2.0

C1 Recovery % 78.2
Cz Recovery % 90.1
Cat Recovery % 97.0
lThe recovery is based on CO2 rejection column being operated for ethane
reflection/propane recovery.
For enhanced oil recovery programs utilizing COz injections,45 membrane

separation technology is also finding increasing application. Carbon dioxide is
injected into the well as a miscible flood to remove additional oil from de-
pleted fields. It has been found that CO2 at concentrations above 90% will sol-
ubilize oil absorbed in the substrata, allowing secondary and tertiary recovery.
In EOR production, CO2 injected into the formation eventually reemerges
mixed with light hydrocarbon gases associated with crude oil. A membrane
system can be used to recover CO2 for reinjection and recover the light hydro-
carbons for sale as pipeline gas.
Goddin presented a detailed cost analysis for recovering CO* from casing-
head gas where bulk removal is effected by membrane separation followed by
a conventional gas treating process. His analysis showed that for feed gas con-
taining more than 30% COz, the process heat requirements for a membrane sep
aration process are approximately 30 to 40% lower than for a cryogenic process.
When compared to an acid gas removal process using DEA as a solvent, the mem-
brane separation process requires only 20 to 25% of the process heat require-
ment.
There are many wells containing about 40 to 80% CO* that could be proc-
essed with a membrane system, providing a high purity COz for EOR and a fuel
gas by-product.
Membrane gas separators may also be used to separate methane from carbon
dioxide formed during the decomposition of organic matter in landfills.64
Permea’s PRISM@ gas separators are being used in a Florence, Alabama landfill
to upgrade raw to 95% methane purity so that it can be routed to the city gas
department.
The Ube system may also be useful for CO2 separations,& but no data have
been reported.
Oxygen/Nitrogen Separation
The U.S. produces approximately 15 millions tons of oxygen from air each
year.66 Most of this production is accomplished cryogenically, and in large plants
the cost is very low. Uses of oxygen in quantities of less than 10 tons per day,
however, constitute a substantial share of this market. It is thought that mem-
brane systems could economically enrich air at this smaller plant size.67
In many applications, such as inert blankets for combustible materials or
chemicals, nitrogen rather than oxygen is the desired separator product. Permea
in conjunction with Maritime Protection S.A. (wholly owned subsidiary of
Permea Kristiansand, Norway) commercialized a system to provide >95% pure
nitrogen aboard ships using PRISM@ separators.6*f69 The system can process up
to 100,000 scfld of air and operates at 120’ to 15O’F and pressures up to 425
psig. At the optimum operating conditions, the system can produce nitrogen at
less than 44 per normal cubic meter or less than $1.40 per thousand scfm.“p@
More recently, membrane systems are planned to provide nitrogen on offshore
platforms and in remote landbased applications. Other possible applications in-
clude nitrogen blanketing of food to prevent spoilage and to provide nitrogen for
manufacturing plant operations.
Permea has under development a new membrane which has approximately
3 times the permeability of the currently used PRISM@ separator and the same
selectivity.‘” This membrane system should provide nitrogen at a significantly

lower cost thus broadening the applicability of membranes for 02/Nz separa-
tions.
The Generon Air Separation Membrane System which was introduced by
Dow Chemical Company in 1985, also is used to provide 95% nitrogen for blan-
keting. The cost per unit volume nitrogen produced is similar to that reported
by Permea.17 One major difference is the Generon system uses 75 to 90 psi air
with an upper limit of 115 psig whereas Permea prefers to use air at 300 to 600
psi. The Permea system operates over a rather broad temperature range and is
quite flexible. The Generon system operates preferably at about 15’ to 25’C
which requires refrigeration of the feed stream. The choice of system will de-
pend largely on the availability of compressed air and the constraints of a given
application. Both systems produce nitrogen with a dew point of about -65°C.
The Asahi Glass system is designed to produce oxygen enriched air of up to
40% for medical applications. The driving force is maintained by a vacuum on
the permeate side of the plate and frame system. Air at atmospheric pressure is
the feed.
Signal’s UOP Fluid Systems Division system also produces oxygen enriched
air of 30% 0s for commercial applications. Japan’s Osaka Gas has developed a
similar system.15 Again, air is supplied to the membrane at atmospheric pressure
and the driving force is maintained by a vacuum on the permeate side of the
membrane. Oxygen enriched air can be supplied at very attractive costs.
Dehydration
A potentially large market for membrane separators exists in dehydration. A
hollow fiber membrane separator system is being developed by Permea Inc. to
dehydrate water/alcohol mixtures including methanol, ethanol, isopropanol,
and/or butanol. The system is not limited to azeotropic mixtures, but it will
eventually be used in the dehydration of a wide variety of water/alcohol mix-
tures.” Although the first system investigated was ethanol/water, the work
aims at separating water/organic mixtures in general.‘l One such application
is the removal of water from natural gas to bring it to pipeline specifications.
The cellulose acetate systems also are useful for natural gas dehydration. A
major problem is the loss of CH4 during dehydration, and it appears that cur-
rently available membranes are not suitable in this regard.” More water per-
meable and more selective systems are required.
Other Separations
Pervaporation is a special case of gas separations in that it is a concentra-
tion driven process. The feed mixture is supplied as a liquid to the membrane
and the permeate is recovered as a vapor on the low pressure side of the mem-
brane. Pervaporation finds application in dehydration as well as the separation
of a variety of liquid mixtures. A discussion of pervaporation is beyond the
scope of this review.
Liquid separations including reverse osmosis, ultra and microfiltration are
pressure driven membrane processes. Membranes are broadly applied to effect
separations in these areas. They are used in the making of potable water, in fil-
tration and concentration in the food industries, in the electronic industries
and find wide application in medicine and health care. As with pervaporation,
these membranes also are beyond the scope of this review. The reader is re-
ferred to an outstanding, up to date discussion of the membranes which are
used for these processes as well as summary of the processesthemselves by R.E.
Kesting.n
ACKNOWLEDGEMENT
We wish to thank Dr. M.A.E. Kraus, now with Gelman Industries, who con-
tributed the introductory membrane and fundamental sections of the text, and
Dr. R.S. Narayan for his support and contributions to the writing of this chapter.
REFERENCES
1. Graham, T., Phil. Mag., 32, 402 (1866).

2. Yasuda, H. and Stannett, V., in Polymer Handbook, D. Brandrup and E.N.
Immergut, eds., John Wiley and Sons, New York, p. 11 l-229 (1975).
3. Loeb, S. and Sourirajan, S., Report No. 6060, Dept. of Engineering, Uni-
versity of California, Los Angeles (1960).
4. Ward, W.J., III, Browall, W.R. and Salemme, R.N., J. Membr. Sci., 1, 99
(1976).
5. Gantzel, P.K. and Merten, U., lnd. Eng. Chem. Process Res. Dev., 9. 331
(1970).
6. Merten, U. and Gantzel, P.K., U.S. Pat. 3,415,038 (1968).
7. Henis, J.M.S. and Tripodi, M.K., U.S. Pat. 4,230,426 (1980).
8. McBride, R.B. and McKinley, D.L., Chem. Eng. Prog., 61, (3) 81 (1965).
9. DuPont Company, Permasep Technical Bulletins 105, 110 (1972); Gardner,
R.J., Crane, R.A. and Hannan, J.F., Chem. Eng. Prog., 73, 76 (1977).
10. Matson, S.L., Lopez, J. and Quinn, J.A., Chem. Eng. Sci., 38, 503 (1983).
11. Chemical Economy and Engineering Review, 16, No. 7-8, p. 37 (1984).
12. Makino, H., Kusuki, Y., Harada, T., Shimazaki, S. and Isida, T., U.S. Pat.
4,440,643 (1984).
13. Chemical Week, 137, p. 39 (July 10, 1985).
14. Chemical Week, 134 (January 19, 1984).
15. Chemical Week, 135, p. 38 (July 24, 1984).
16. Membrane Separation Technology News, 3, p. 1 ( 1985).
17. McReynolds, K.B., Chem. Eng. Prog., 81, 27 (1985); Industrial Chemical
News, 6 (April 1985).
18. Zampini, A. and Malon, R.F., Proc. ACS Div. Polym. Mat., Sci. & Eng., 52
345 (1985).
19. Hwang, S. and Kammermeyer, K., Membranes in Separations, John Wiley
and Sons, New York, p. 501 (1975).
20. Reference, 19, p. 506.
21. Perry, R.H. and Chilton, C.H., Chemical Engineers’.Handbook, McGraw-Hill
Co., New York, p. 17-45 (1973).
22. Vieth, W.R., Howell, J.M. and Hsieh, J.H.,./. Membr. Sci., 1, 122 (1976).
23. Stern, S.A. and Frisch, M.L., Ann. Rev. Mater. Sci., 11, 523 (1981).
24. Masuda, T., Isobe, E. and Migashimuna, T., J. Am. Chem. Sot., 105, 7473
(1983).
25. Paul, D.R., Ber. Bunsenges. Phys. Chem., 83, 294 (1979).
26. Koros, W.J. and Paul, D.R., J. Polym. Sci., Polym. Phys. Ed., 16, 1947
(1978).
27. Raucher, D. and Sefcik, M., in Industrial Gas Separations, T.E. Whyte, Jr.,
C.M. Yon and E.H. Wagener, eds., ACS Symposium series 223, American
Chemical Society, Washington, D.C., p. 111 (1983).
28. Pace, R.T. and Datyner, A., J. Polym. Sci., Polym. Phys. Ed., 17, 437
(1979).
29. Brandt, W.W.,J. Phys. Chem., 63, 1080 (1959).
30. DiBenedetto, A.T., J. Polym. Sci., Al, 3477 (1963).
31. Weller, S. and Steiner, W.A., J. Appl. Phys., 21, 279 (1950).
32. Weller, S. and Steiner, W.A., Chem. Eng. Prog., 46, 585 (1950).
33. McLean, D.L., Stookey, D.J. and Metzger, T.R., Hydrocarbon Process.,
62 (8). 47 (1983).
34. Henis, J.M.S. and Tripodi, M.K.,J. Membr. Sci., 8, 233 (1981).
35. Fritzsche, A.K., Holladay, H.P. and Woodcock, M.L., U.S. Pat. 4,323,453
(19821.
36. Zampini, A., U.S. Pat. 4,323,453 (1982).
37. Vos, K.D. and Burris, F.O., Jr., lnd. Eng. Chem. Prod. Res. Dev., 8, 84
(1969).
38. Agrawal, J.P. and Sourirajan, S., J. Appl. Polym. Sci., 14, 1303 (1970).
39. McDonald, W. and Pan, C., U.S. Pat. 3,842,515 (1974).
40. Manos, P., U.S. Pat. 4,080,743 (1978).
41. Schell, W.J. and Houston, C.D., Hydrocarbon Process., p. 249 (September
1982).
42. Schell, W.J., Hydrocarbon Process., p. 43 (August 1983).
43. Schell, W.J., J. Membr. Sci., 22, 217 (1985).
44. Makino, H., Kusuki, Y., Harada, T. and Shimazaki, H.,U.S. Pat. 4.378.400
(1983).
45. Nakamura, A., Energy and Resources, 6, #3, 67 (I 985).
46. Nakamura, A. and Hotta, M., Kagaku Keizai, pp. 13-19 (May 1985).
47. Erb, A.J. and Paul, D.R., J. Membr. Sci., 8, 11 (1981).
48. Yasuda, H. and Rosengren, K., J. Appl. Polym. Sci., 14, 2839 (1970).
49. Koros, W.J., Separations Research Program, Fall Meeting (October 28-29,
1985).
50. Dyer, B.S., Day, M.R. and Bergenn, W.R., Encyclopedia of Polymer Science
and Technology, John Wiley and Sons, NY, Vol. 9, pp. 450-459 (1968).
51. Paul, D.R. and Morel, G., Kirk-Othmer Encyclopedia of Chemical Tech-
nology, John Wiley and Sons, NY, Third Edition, Vol. 15, p. 118 (1981).
52. Steigelman, E.F., U.S. Pat. 4,014,665 (1977).
53. MacLean, D.L., Prince, C.E. and Chae, Y.C., Chem. Eng. Prog., 76, 98
(1980).
54. Shirley, J. and Coreik, D., Chemical Processing, 30 (Jan. 1982).
55. Bollinger, W.A., Long, S.P. and Metzger, T.R., Chem. Eng. Prog., 80, 51
(1984).
56. MacLean, D.L., Bollinger, W.A., King, D.E., Narayan, R.S., Paper presented
ACS meeting (Aug. 1981).
57. Bollinger, W.A., MacLean, D.L. and Narayan, R.S., Chem. Eng. Prog., 78,
27 (1982).
58. Schell, W.J., Lawrence, R.W. and King, W.M., Membrane Applications to
Fuel Conversion Processes, Final Report to the Energy Research and
Development Administration, Report No. RE-2000-4 (1976).
59. Cooley, T.E. and Coady, A.B., U.S. Pat. 4,130,403 (1978).
60. Cooley, T.E. and Dethloff, W.L., Chem. Eng. Prog., 81, 45 (1985).
61. Boustany, K., Narayan, R.S., Patton, C.J. and Stookey, D.J., paper pre-
sented GAP National meeting (March 1983).
62. Waste Age, pp. 28-30 (November 1983).
63. Goddin, C.S., Hydrocarbon Process., p. 125 (May 1982).
64. Mazur, W.H. and Chan, M.C., Chem. Eng. Prog., 78, 38 (1982).
65, Enhanced Oil Recovery, H.K. Van Poolen and Associates, inc., Pennwell
Publishing Company, pp. 132-141 (1980).
66. Follwell, W.F., Chem. Eng. News, 7 (July 15, 1974).
67. Schell, W.J. and Hoenschemeyer, D.L., Paper presented AlChE Symposium
(June 1982).
68. Chemical Engineering, p. 10 (October 17, 1983).
69. Metzger, T.R., Handermann, A.C. and Stookey, D.J., Plant/Operations
Prog.., 4, 168 (1985).
70. Membrane and Separation Technology News, 1, p. 1 (July 1983).
71. Chem. Eng. News, p. 22 (June 27, 1983).
72. Fournie, F. and Agostini, J.P., Proc. of the Offshore Technology Confer-
ence, Houston, TX, pp. 113-121 (1984).
73. Kesting, R.E., Synthetic Polymeric Membranes, Second Edition, John Wiley
and Sons, Inc., NY, Chapter 2 (1985).
While this information is presented in good faith and believed

to be accurate, Monsanto Company does not guarantee satisfac-
tory results from reliance upon such information and disclaims
all liability for any loss or damage arising out of its use and
makes no express or implied representation or warranty as to the
fitness, merchantability or any other matter with respect to the
products or processes referred to herein. Nothing contained
herein is to be construed as a recommendation to use any prod-
uct in process in conflict with any patent.
Index
"Absolute" membranes - 78 Anisotropy - 94-95

"Absolute" retention - 81, 219 Annealing - 35-36
Absorption - 574 Arrhenius equation - 412
Acid dosing - 374 Asymmetric hollow fiber - 149, 151
Activated sludge reactor - 254 Asymmetric membranes - 2, 4, 12-
Activated transport - 403 14, 30, 32, 35, 45, 54, 138,
Activation energy - 401, 435, 351-352, 404, 445, 450-451,
570 455-456, 458, 462, 579
Active charcoal filtration - 376 Augmented cross-flow effects -
"Active" transport - 3 192-195
Adsorption - 402, 574 Autoclave - 116
Adsorption losses - 159-160 Automotive industry - 382-384
Aerosol retention - 86-90
Alamine@ - 336, 521; 528-531, Back diffusion - 413-414
539, 543 Back-diffusive transport - 173,
Amination - 43 186, 195
Ammonia plants - 578, 582 Backflush reactor - 250
Ammonia purge gas - 583-584 Back-wash fluid - 206
Ammonia recovery - 547 Backwashing - 99-101, 118
Ammonia synthesis - 387-388 Bacteria - 61-62, 71, 78-81, 158
Anion-exchange membranes - 40, Bacteria retention - 78-81
43, 499 Batch fermentation - 402, 474
Anisotropic membranes - 106, Batch fermentor - 253
138, 140, 145, 166 Batch operating mode - 216
594
Index 595
Batch reactors - 422 Casting process - 64

Batch recirculation - 490 Casting solution - 17, 28, 30-32,
Beer - 240-241 35, 146
BET adsorption data - 78 Casting tube - 144
Beta ratio - 71 Cathode rinse stream (or
Biomass - 467-468, 470-471 catholyte) - 498-499
Biomembrane plants - 255 Cation-exchange membranes - 40,
Bipolar membranes - 490-496 42, 499
Blood - 458, 462, 492-493 Caustic extraction effluent-
Blood plasma processing - 242- color rejection - 232
243- Cell concentrations - 470, 472
Boundary layer conditions - 503 Cell harvesting/washing - 128-
Boundary layer thickness - 415, 131, 242
417 Cell pair - 487
Brackish water membrane - 267 Cell recycle fermentors - 470-472
Bubble-point - 78-81, 158 Cellular structure - 21, 23, 34
Bubble-point equation - 72 Cellulose acetate - 23-25, 36,
"Bubble-point" pressure - 116 46, 309, 578-580
Bubble-point test - 61, 71-76 Cellulose acetate membranes -
Bubble-pointable cartridge - 270-271, 273, 309-311, 455,
69, 111 560-561, 586, 588
Bubble-pointable tube - 109 Cellulose ester - 588
Bubble-pointed unit - 106 Cellulose nitrate membrane - 9
Cellulose triacetate - 310, 588
Capillary membranes - 53-54, Centrifugal force - 192
421 Ceramic membranes - 4, 5, 105,
Capillary module - 50-53 115, 154-155
Capillary-pore membranes - 7-9, Challenge tests - 70-71
62-64, 66-69, 78, 82-83, "Charged" membranes - 161, 201,
90, 98-99, 106, 108 220
Carbon dioxide separation - Cheese whey protein recovery -
588-589 232-236
Carnell-Cassidy technique - 310 Chloralkali cells - 483-484
Cartridge filtration - 375 Chlorine dosing - 374
Cartridge-type filters - 50 Chlorine resistance - 324
Cascade without reflux - 364 Cleaning by recycling - 206-207
Cascades - 360-373 Cleaning in place - 491
Casein - 492 Coating - 575, 577
Casing-head gas - 589 Cobalt recovery - 542-543
CaS04 precipitation - 507 Co-coupled transport (or co-
Cassette plate and frame system - transport) - 516-517
209, 211 Coenzymes - 424-425
Casting anisotropic ultrafil- Cofactor - 403, 409, 453-454
tration membranes - 140 Co-ions - 40, 483
Casting machine - 54 Collapse of pores - 143
Collodion membranes - 61 history - 512-515

Colloidal particles - 189 theory - 517-520
Colloidal suspensions - 180-186 Coupling agent - 403, 456
Compaction - 312 Covalent binding - 403, 456, 458
Complexing agents - 526-531, 537- Critical voltage - 195
541 Cross-flow electrofiltration -
Composite membranes - 45-49, 271, 193-195
273, 560-561, 574-579 Cross-flow filtration - 99, 102-
Compression - 574 106, 109, 112, 124-134,
Concentration - 3, 37 382-384
Concentration difference - 3 Cross-flow membrane cassette -
Concentration effects - 521-523 108
Concentration gradient - 3 Cross-flow process - 268
Concentration polarization - 163- Cross-flow ultrafiltration
164, 166-198, 268-269, 349- system - 171
351, 403, 410-411, 413-415, Cross-flow velocity - 199
417-419, 421, 426, 428, Crosslinked PPO (polyphenylene
465, 472, 502-507, 532-537 oxide) - 580
Concentration profiles - 29, 31- Crosslinking - 456, 483-484
32, 349, 352, 357, 372-373 Crown ethers - 517, 540-541
Conductance - 504 Cryogenic process - 588-589
Confinement - 402 Cryogenics - 574
Constant reflux - 364
Contact angle - 72, 77 Dairy applications - 232-237
Continuous cell culture - 131 Dead-end units - 410, 421
Continuous fermentor - 253 Dead-ended filtration - see
Continuous flow reactors - 402 "Through-flow filtration"
Convection - 435, 437 Dehydration - 590
Conversion - 404, 410-411, 417- Dehydration of sour gases - 389-
418, 422, 426, 435, 437- 390
438, 440-441, 444-448, 450- Deionized water - 119-120, 123, 218
451, 454, 456-457, 459, Density of pores - see Porosity
461-462, 465-466, 474, 476 Desalination membranes - 560
Copolymerization - 403 Desalination of brackish water -
Copper etchant baths - 545-546 376-382
Copper recovery - 541-542, 545- Desalination of seawater - 299-
546 302, 336, 374-376, 490-491
Corn starch - 241 Desalting of seawater - 491
Counter-coupled transport (or Detergent cleaning - 199
counter transport - 515-516 Diafiltration - 129, 243-247,
Counter ions - 40, 483 418, 422
Coupled transport membranes - Dialysis - 1
511-558 Dialysis membranes - 410, 445,
characteristics - 520-537 467, 474
Index 597
Diffusion - 435, 437, 444-448, Electrophoretic deposition - 222

450-451, 455, 459, 461- Electrophoretic transport - 195
462, 570 Electroplating - 493
Diffusional deposition - 88-90 Electroplating rinse waters -
Diffusional flow - 74 543-545
Diffusional resistance - 409- Electrostatic attraction - 84
410, 446, 450, 457-458 Emulsion liquid membrane process
Diffusivity - 3, 37, 564-565, design - 552-555
569 Emulsion membranes - 514-515,
Dip-coating - 46 520
Direct interception - 88 Encapsulation - 402
Dirt-loading capacity - 90, 93 End seal - 579-581
Disperse soluble immobilized Enhanced oil recovery - 582,
enzyme - 251 588-589
Donnan exclusion - 161, 483, Entrapment - 404, 455
494, 503 Enzyme activity - 409, 411, 417,
Donnan ion effect - 322, 339 422, 424, 444-445, 451,
Driving forces - 2-4 462, 464
Drum-casting machine - 140-141 Enzyme concentration and purifi-
Drying - 560, 578-579 cation - 242
Dual sorption theory - 566, 568 Enzyme deactivation - 415, 417,
Dyestuff recovery and purifi- 419, 440, 445
cation - 229 Enzyme depth of penetration -
Dynamic membranes - 153-154, 281 427, 434
Dynamic secondary membrane - Enzyme engineering - 401
164, 167 Enzyme half-life - 411-413, 422
Enzyme membrane reactors - 403-
EDR process - 498, 507-508 404, 409-426, 428, 465
EDTA-Cu complex - 494 Enzyme reactors - 247-252
Egg white - 237-238 Enzyme stability - 404, 411,
Electric field - 192-195 427, 464
Electrical potential difference Equilibrium curve - 363, 367
-3 Equilibrium state - 19
Electrical resistance - 486, 488 Evaporation - 14, 16, 382-384
Electrocoat paint - 222-224 External mass transfer resist-
Electrode isolation - 498-499 ances - 435
Electrode reaction - 496-498 Extractive distillation - 396-
Electrodes - 496-502 398
Electrodialysis - 1, 482-510
applications - 490-494 "Facilitated" transport - 2-3,
costs - 509 38
Electrodialysis stack - 487-490 "Fast" gases - 582
Electrokinetic adsorption - 83 FDA approval - 330
Electrolytic cells - 482-483 Feed and bleed - 490
Feed and blend mode - 216 Gel concentration - 426, 428

Feed composition - 573 Gel formation - 426-428, 434,
Feed pretreatment - 374 438-439
Feedwater analysis - 282 Gel layer - 169-170, 186,
Feedwater pH - 284, 289 383-384, 403-404, 418,
Fermentation - 466 426-428, 434-436, 438,
Fermentation efficiency - 470 465
Fermentor productivity - 466, Gel polarization - 164, 167-174,
471-472, 474 196
Fick's law - 3, 18, 518-519, Gel-polarization model - 167
564, 569-570 Gelatin - 238
Filtration properties - 25, 27 Gelation time - 23, 27
Finger-type structure - 20-21, Glassy polymers - 565-570
25, 27, 33-34 Glycerin - 28
Flat membranes - 410, 419, 434- Grow-through - 78
435, 437, 465
Flat-plate cartridge - 210 Heavy metals - 124
Flat-plate membrane - 210 Heavy metals removal - 230
Flat sheet membranes - 37, 50, Henry's law - 564, 566
53-54 Henry-type sites - 566-569
Float casting - 309 Heterogeneous membranes - 2, 44,
Flow permeability test - 77 484-485
Flue gas - 496 Hollow fiber fermentor - 473
Fluid polymer layer - 30 Hollow fiber module - 50-52,
Flux decay - 104, 198-202 354-360, 380
Flux variation - 172 Hollow fibers - 37, 53-55, 205-
Fluxes - 2-4 207, 214, 357, 404, 410,
"Forward-flow" test - 74-75 421, 440, 445, 450-451,
Fouling resistance - 150 454, 458-459, 463, 467,
Fractionation capability - 86 473, 547-550, 561, 581,
Fruit juice - 238-240 586, 588
FT-30 composite membrane - 327- Hollow fine fiber packaging -
331 279-281, 285
Functional synthetic membranes Homogeneous membranes - 2, 4,
56 37-45, 486
Furfuryl alcohol - 334 Homogeneous metal and glass
membranes - 37-38
Galvanic industry - 384-387 Homogeneous polymer membranes -
Gas permeation - 351-352, 357, 37
359, 363 Hydrocrackers - 585-587
Gas-phase deposition - 46-47 Hydrogen - 38
Gas separation - 12, 37, 360 Hydrogen recovery - 387-389,
Gasket spacer - 487 583-588
Gel (cake) resistance - 168 Hydrogen sulfide - 588
index 599
Hydrolysis - 43, 418, 421, 426, Ion exchange demineralization -

440, 450, 454, 467, 475- 290, 296-298
476 Ion exchange membranes - 39-45,
Hydrophilic polymers - 160 483-484, 486-487
Hydrophobic membrane - 116 Ionic binding - 456
Hydrophobic polymers - 160 Ionic velocities - 482
Hydroprocessors - 582 Ionics, Inc. - 488, 508
Hydrostatic pressure - 3 Isoelectric point - 173-174, 492
Hydrotreaters - 586 Isotopes - 560
Immobilization techniques - Kelex@ 100 - 521, 524-525, 527,

402-403, 409, 438, 529, 532, 534-535, 539,
440, 455 542-543, 550
Immobilized cells - 254 Kermac@ 470B - 524, 527, 529,
Immobilized enzyme batch 534
membrane reactor - 428- Knudsen flow - 562, 576
429, 432
Immobilized enzymes - 249, Lactose - 234
251, 402-403, 418, 426, Laminar flow - 174-176, 177,
434, 437-438, 450, 456, 350, 438, 441, 447, 452,
458, 462, 465 457-458
Imperfections - 577 Landfills - 589
Industrial laundry waste- Langmuir constant - 568
water - 125-126 Langmuir-type sites - 566-569
Industrial reverse osmosis Latex concentration/recovery -
at a refinery - 290-296 229-230
Industrial scale membrane Lime clarification - 290, 292-293
production - 49-53 Limiting current density - 502,
Industrial wastes - 303 506-507
Industrial water - 302-303 Liquid films - 4
Inertial impaction - 88 Liquid membranes - 38-39
Inhibitors - 410, 418, 440, LIX@ 54 - 521, 525, 527, 532,
454, 473-474 534, 542-543
Inorganic coatings - 201 LIX@ 64N - 521-523, 525-529,
Inorganic materials - 11-12 532, 538, 554
Inorganic membranes - 153-155 Lock, Sidney - 137, 139, 270
Integrated circuits - 491-492 Loeb-Sourirajan membrane - 137,
Interaction parameter - 570 139, 309, 311-312, 560, 575
Interfacial membrane formation, Long-term flux decay - 199
mechanism - 332-333
Interfacial polymerization - Macromolecular substrates - 409-
46-48 410, 418, 426
Intrusion pressures - 77 Manufacturing equipment - 53
Ion exchange, reverse osmosis Mass transfer coefficient - 174-
and - 296-298 179, 349, 435
600 Handbookof industrial
Membrane Technology
Mass transfer diffusional Michaelis-Menten rate equation -

resistance - 445, 450, 442, 447
456-457 Microbial cells - 402, 455, 474,
Media migration - 91 476
Membrane barrier layers - 272 Microelectronics - 120
Membrane column - 360, 368-373 Microfiltration - 5, 50, 61-135,
Membrane configuration - 106- 383-384, 439
114, 202-214 applications - 114-134
Membrane (definition) - 2 Microorganisms - 11, 70-71, 116,
Membrane degradation - 551 401-402, 451, 455, 468
Membrane distillation - 6-7, Microporous membranes - 4-36,
132-134 466, 559-561
Membrane electrodes - 457 Microporous metal membranes - 12
Membrane fermentors - 252-254, Microporous polysulfone -
404, 466-476 311-313, 333, 343
Membrane flux - 166-198, 288, Microporous sintered membrane -
523 4-5
Membrane fouling - 198-202, 224, Microporous support - 48-49
285, 327, 418, 465, 472, Milk - 492
507-508 Milk concentration - 236-237
Membrane modules - 50-53 Miscibility gap - 15-17
Membrane plugging - 90-106 Mobile cells - 252-254
Membrane preparation - 4-56 Mobile enzyme - 248-249
Membrane properties - 486-487 Molecular weight cut-off - 155-
Membrane reactor - 249 156
Membrane rejection - 413 Municipal potable water - 302
Membrane structure - 20-28, 62- Municipal wastes - 303
64, 138-155
Membrane surface treatments - Natural gas - 588
200-202 Nernst idealization - 503
Membrane testing and evalu- NF-40 composite membrane - 323-
ation - 582 324
Meniscus coating - 310 NF-50 composite membrane - 332
Mercury intrusion test - 77 Nickel plating - 385-386
Mercury porosimetry data - 81 Nickel recovery - 542-543
Metal alloys - 11 Nitrocellulose - 61
Metal cleaning - 225 Nodular structure - 20-21, 23,
Metal finishing - 493 34
Metal ion concentration Nonpolar gases - 560
effects - 523-524 Nonporous membranes - 560-561,
Metal ion effects - 525-526 563-565
Methanol synthesis purge - Nonsolvent addition - 15, 17
584-585 NS-100 membrane - 314-316
Michaelis-Menten model - 411, NS-200 membrane - 333-335
421, 437 NS-300 membrane - 320-323
Index 601
NTR-7250 composite membrane - Phenol - 42

324-326 Phenol recovery - 547
Photographic solutions - 494-495
Oil fields - 582 Piperazine - 320-323
Oil-water separation - 224-226 Plasma polymerization - 47, 340
Oily wastewaters - 226, 384 Plasmapheresis - 126-128
Operating line - 363 Plasticizers - 142
Optimum recirculation rate - Plate and frame configuration -
216-217 274-275
Organic rejection - 317 Plate and frame module - 50-51,
Osmotic pressure model - 166,- 581
167 Plate and frame units - 106-
Oxygen enrichment - 581 109, 208-211, 214
Oxygen/nitrogen separation - Pleated cartridges - 50-51, 106,
589-590 109-110
Point-of-use processing - 221
PA-300 membrane - 316-318 Poiseuille's law - 452-453, 562,
Parasite drag - 204-205 586
"Passive" transport - 2 Polar gases - 560
PEC-1000 membrane - 335-338 Polarization layer - 410, 417
Peptides - 402 Pollution control, reverse
Permeability - 3, 563-566, 569- osmosis and - 298-299
571, 574-576, 578, 580- Polyacrylonitrile - 341
581 Polyamide - 23-27, 271, 314
Permeate - 351, 353, 356, 362, Polybenzimidazolone membrane -
391, 410-411, 415, 417- 342-343
418, 421-422, 425, 428, Polycondensation - 42
434, 438, 465-466, 471- Poly(dially1 amine) - 319
473 Polyelectrolyte complex ultra-
Permselectivity - 40, 487 filtration membranes - 161
Pervaporation - 37, 390-398, Polyepiamine - 316
590-591 Polyepichlorohydrin - 316, 318
Pesticide removal - 316 Polyethylene - 10, 41
pH and gel polarization - 173- Polyethylenimine - 48, 314
174 Polyimide - 579-581, 586
pH effects - 525-526 Polyimide composite membrane -
Phase-inversion membranes - 342
9-12, 20-28 Polyimide resin - 561
Phase-inversion process - 12, Polymer concentration - 23-27,
64 29, 35
Phase separation, mathematical Polymer content additives -
description of - 18-20 146-147
Phase separation process - Polymerization - 42
14-15 Poly(4-methyl pentene-1) - 580-581
Polyolefin - 561, 581 Product flow as function of

Polyphenylene oxide (PPO) - temperature - 266
561, 581 Product-inhibited enzymes -
Polyphosphate dosing - 375 403, 450
Polypropylene - 10 Product water flow - 264
Polystyrene - 41, 567 Proteins - 493
Poly(styrenesulfonic acid) - Pulp and paper waste treatment -
343 230-232
Polysulfone - 9, 23-25, 41, Purity - 570, 572-573
43, 47-48, 455-456, 462, Pyrogens - 84, 142, 221
465, 561, 575-576, 579-
581, 586 Quaternization - 43
Polytetrafluoroethylene - 4,
44 RAD wastes - 124
Polyurea - 314 RC-100 membrane - 316-318
Polyvinyl alcohol - 326 Reactor capacity - 422
Polyvinyl alcohol recovery - Reactor productivity - 421-422,
227-229 437
Poly(4-vinylphenol) - 343 Recarbonation - 293
"Pore-flow" model - 161-163 Recovery - 570, 572
"Pore-former" - 64-65 Recovery of reverse osmosis
Pore size - 4-9, 12, 61, 73, plant - 267
155-156, 158 Rectifier - 500
Pore size determination - 70- Recycle reactor - 250
78, 155-158 Reflux cascades - 364-366, 394
Porosity - 6, 23, 25, 49, 64, Rejection of sodium chloride - 267
69, 75, 77, 145, 148 Resistivity of deionized
Porous support - 45 water - 120, 122
Posttreatment design - 290 Retention - 164-165
Power supplies - 500-502 Retention by adsorption - 82-83
Pre- and post-precipitation Retention by "charged" mem-
procedures - 35-36 branes - 83-86
Precipitant - 14, 27, 29, Retention characteristics -
31-32, 34 78-90, 159-166
Precipitation - 14, 25-28, 30 Retention of deformable
Preferential sorption-capil- particles - 82
lary flow mechanism - Retention of spherical and
264 linear molecules - 157-158
Prefilters - 90-94, 198 Retentivity - 156-158
Prefiltration - 131 Reverse osmosis - 1, 20, 37, 119,
Pressure effect - 200 136, 233, 260-306, 352, 356,
Pretreatment - 508 359, 374-382, 384-387, 403,
Pretreatment design - 283-285 424, 450
Index 603
applications - 260-262, Sodium hexametaphosphate - 507-

302-303 508
costs - 303-304 Solid polymer layer - 30
design - 286-290 Solrox 0200 membrane - 340-342
future projections - 305 Soluble enzyme batch membrane
plant design - 281-290 reactor - 428-429, 432
process considerations - Solubility - 565, 568-569
263-270 Solubility parameter - 148
Solubility parameter disparity -
Salt flow - 265-266, 269 27-28
Salt rejection - 35-36 Solution coating technique - 272
Sanitizing agents - 220 'Solution-diffusion" model -
Seawater desalination - 299-302, 161-162, 264-265
336, 374-376, 490-491 Sorption enthalpies - 568
Seawater membrane - 267 Sour gas purification - 588
Secondary flow effects - 177 Sourirajan - 137, 139
Seeding technique - 382 Soy whey - 237
Selective barrier - 45 Spacers - 487, 489
Selectivity - 362, 570, 573- Spherical cells - 20, 22
575, 577, 579 Spinneret - 54-55, 151
Semiconductor industry - 218- Spinning hollow fibers - 149-153
221 Spiral wound configuration - 274-
Semiconductor process fluids - 278, 285
119-124 Spiral wound modules - 50-51, 212-
Separation factors - 563, 580 214, 379, 561, 581
Separation of gases - 559-593 Sponge ball cleaning - 203
Separation unit - 361 Sponge-type structure - 20-22,
Shape-factor - 80 25, 33, 35
Shear-stresses - 418, 426, 434, Stable state - 19
438, 441, 445 Stack power - 496-502
Sheet-flow stack - 488 Stage - 361
Shrinkage - 30, 33 Steam sterilization - 116
Sieve mechanism - 264 Sterilization - 114-119
Silica rejection - 329 Stokes-Einstein relation - 181,
Silicone - 575-576, 580-581 185, 530-531
Silt density index - 282-284 Stretched membranes - 6-7
Silver - 494 Stretching-process - 64-65
Skin - 20-21, 33-34 Sulfochlorination - 43
Skin-type membranes - 14, 32-33 Sulfonated polyphenylene oxide
"Slow" gases - 582 membranes - 339
Smoke DOP test - 78 Sulfonated polysulfone
Sodium base spent sulfite membranes - 338-339
liquor - 231 Supported liquid membranes -
Sodium bisulfite dosing - 376 38, 514, 520, 547-551
Surface texture - 328 Ultrafiltration - 5, 53, 136-259,

Symmetric membranes - 2, 4, 12- 382-384, 403, 410, 421, 426,
13, 21, 30, 404, 458, 462 428, 434, 438-440, 445, 450-
453, 455, 465-466, 476, 492
"Tapered design" - 376, 378 operating modes - 214-216
Temperature - 565 Ultrapure water - 8, 121, 218-221
Temperature effect - 195-198 Ultrasonic welding process - 54
Ternary phase diagram - 16-17, Ultraviolet sterilization - 122
145, 147 Unit-cell stack - 488-490
Tetrafluoroethylene - 6 Univalent-cation-selective
Thermal gelation (thermo- membrane - 499
gelation) - 10, 14-16 Unstable state - 19
Thermal-phase-inversion Unsupported liquid membranes - 38-39
process - 65-66 UOP Butamer process - 389
Thiele modulus - 435-437, 444- Uranium recovery - 543-544
445, 448-450, 459-461 U-tube liquid membrane - 512-513
Thin film composite reverse
osmosis membranes - 307- Vent filters - 116
348 Vexare spacer - 488-489
Thin film polymerization - 272 Viable cells - 466-467, 470-473
Threshold pressure - 170, 200 Virus concentration - 247
Through-flow filtration - 99, Volatile solvent - 16
102, 114-124, 131 VTE (vertical tube evaporator)
Throughput - 90, 95-98, 112 seeding process - 382
Time lag - 564, 569
Tortuosity factor - 78 Waste lubricating oil - 226-227
Tortuous-path spacer - 488-489 Waste treatment - 254-255
"Tortuous-pore" membranes - 62- Wastewater treatment - 382
66, 82-83, 86, 90, 98-99, Water - 24-26, 28, 31
103, 106, 109, 111, 113 Water clustering concept - 264
"Track-etch" process - 7, 66-67 Water transport - 487
Transport number - 487, 503 Watts-nickel bath - 385, 387
Transport rate - 571 Well-head gas - 582
Trimellitic anhydride acid Wet-spinning process - 54-55
chloride - 330 Wetted surface mechanism - 264
Trimesoyl chloride - 320-321 Whey - 324, 473, 491
Tubular configuration - 214, Whey protein concentrate - 235
277-279, 285 Whole cells - 401, 404, 455
Tubular membranes - 53-54, 434, Wine - 240
437-438
Tubular module - 50-52, 354 Yield - 362
Tubular pinch effect - 186-192
Turbulent flow - 176-178, 184 Zr(IV) oxide - 154

Handbook of Industrial Membrane Technology

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Handbook of Industrial Membrane Technology

Transféré par

Droits d'auteur :

Formats disponibles

HANDBOOK OF

Published in the United States of America by

Library of Congress Cataloging-in-Publication Data

Handbook of industrial membrane technology

Richard W. Baker Gabriele lorio

A. Keith Fritzsche Richard G. Sudak

To the best of the Publisher’s knowledge the

Final determination of the suitability of

1. SYNTHETIC MEMBRANES AND THEIR PREPARATION .......... .I

Ion-Exchange Membranes. ....................... .39

2. MICROFILTRATION .................................. .61

Sterilization and Particle Removal (Beverages) .......... 117

3. ULTRAFILTRATION. ................................ .I36

Centrifugal Force ............................. 192

Summary. ..................................... .255

4. REVERSE OSMOSIS. ................................. .260

5. THIN FILM COMPOSITE REVERSE OSMOSIS MEMBRANES. ..... .307

6. PROCESS DESIGN AND OPTlMlZATlON ................... .349

7. ENZYME MEMBRANE REACTORS AND MEMBRANE

Complex Fermentation Systems. .................... .475

8. ELECTRODIALYSIS. ................................. .482

9. COUPLED TRANSPORT MEMBRANES. .................... .511

Supported Liquid Membranes Process Design ............ 547

10. THE SEPARATION OF GASES BY MEMBRANES. ............ .559

This handbook emphasizes the use of synthetic membranes for separa-

o,oor)l UM 0.001 !JM 0,1 FIM 10 UM

Figure P.l: Pore sizes of RO, UF, and MF membranes.

1000 Vitamin 812 -

Water _ Reverse osmosis

Figure P.2: Typical species retained by MF, UF and RO membranes.

Again, the dividing line between UF and RO membranes at 0.001 microns

Figure P.3: Osmosis.

Figure P.4: Reverse osmosis.

Figure P.5: Variation of osmotic pressure with salt concentration.

It has been estimated that the current worldwide membrane/equipment

Pleasanton, California M.C. Porter

Synthetic Membranes and

In recent years, membranes and membrane separation techniques have grown

gain an understanding of the significance of the various structures used in

Fluxes and Driving Forces in Membrane Separation Processes

therefore, is a special form of the passive transport. Completely different, how-

(a) A hydrostatic pressure difference between two phases separated by

The permeabilities of different components in a membrane depend on the

eral, permeabilities are significantly higher than in diffusion-controlled membrane

DISCUSSION OF TECHNICAL RELEVANT SYNTHETIC MEMBRANES AND

Although synthetic membranes show a large variety in their physical struc-

Neutral Microporous Membranes

Table 1.1: Microporous Membranes, Their Preparation and Application

Membrane type Membrane material Pore size Manufacturing process Application

Ceramic, metal 0.1 -20 pm pressIng and 51nterlng Mlcrollltration

or polymer powder of powder

membrane Homogeneous polymer 0.' -10 I'm stretching of extruded Microfiltration,

Homogeneous polymer 0.02 -10 I'm track-etching Microfiltration

Figure 1.1: SEM of a microporous sintered membrane prepared from a PTF E-

Stretched Membranes. Another relatively simple procedure for preparing

Figure 1.2: SEM of a microporous membrane prepared by stretching an extruded

Figure 1.3: Schematic diagram illustrating the preparation of capillary pore

Figure 1.4: SEM of the surface of a capillary pore polycarbonate membrane.

Figure 1.5: SEM of asbestos filter accumulated on the surface of a capillary

Figure 1.6: SEM of the surface of a microporous cellulose nitrate membrane

Polypropylene or polyethylene can also be used for the preparation of mi-

Figure 1.7: SEM of the surface of a microporous polypropylene membrane

The symmetric, microporous polymer membranes made by phase inversion